An Architecture and Realisation of a Smart City Digital Twin With eHealth Case Studies

Abstract

ABSTRACT

Digital twinning is an advanced technology that involves creating virtual replicas of various physical systems. In smart cities, digital twins serve as digital representations that model and simulate various urban elements, such as environment protection (e.g., air quality), critical infrastructure, transportation networks and other urban management processes. It has recently gained considerable attention for its transformative potential, enabling city authorities to visualise and analyse complex city dynamics for better‐informed decision‐making. Therefore, this paper proposes a simplified layered architecture for smart city digital twins. The layers of the proposed architecture cover the range of operations required by the functionality of the digital twin and the interaction between them, from data transfer or synthesis to big data streaming and intelligent analytics. The paper also introduces an open‐source software tool that realises the proposed architecture, with each layer designed as an independent Python module for easy integration and maintenance. Three case studies are used to demonstrate the capabilities of the tool. One use case addresses short‐term forecasting of the air quality index, whereas the other use case targets the detection of an individual's respiratory condition based on data received from wearable devices. The third case combines the other two cases to offer a warning system for residents with medical conditions based on air quality. The results of the case studies show the tool's ability to effectively handle environment and eHealth‐related use cases and combine them for the welfare of smart city residents, leading to a more resilient health‐focused urban landscape.

Introduction

The rapid growth of urban populations, coupled with increasing demands for sustainable development, has pushed cities to explore advanced technologies to address complex challenges in urban governance. By 2050, it is projected that $$68\%$$ of the world's population will reside in urban areas, increasing pressure on governments to find innovative solutions for issues, such as traffic congestion, energy distribution and environmental degradation, as highlighted by Deng et al. [1]. In response, cities are increasingly leveraging digital transformation technologies to enhance urban governance. Among these technologies, digital twin (DT) technology has emerged as a crucial tool, promising to revolutionise urban management through real-time data insights and predictive capabilities as noted by Ivanov et al. [2].

The concept of smart city (SC) digital twins represents a transformative approach to urban management and planning, leveraging advanced technologies to create virtual replicas of physical urban environments. A DT is defined as a digital representation of a physical entity, which facilitates real-time monitoring, analysis and optimisation of urban systems and processes, as discussed by Mukhacheva et al. [3]. Botín-Sanabria et al. [4] further demonstrate how DTs can serve as adaptive data-rich platforms that enable dynamic and proactive management of urban environments.

This innovative framework serves as a bridge between the physical and digital realms, enabling city planners, policymakers and stakeholders to visualise and simulate various urban scenarios, thereby enhancing decision-making capabilities, as demonstrated by Campo et al. [5].

The integration of big data and the Internet of Things (IoT) is pivotal in the development of DTs, as these technologies provide the necessary data streams that feed into the digital models, ensuring their accuracy and relevance, as described by Mukhacheva et al. [3]. Shen [6] add to this by emphasising how seamless data fusion from IoT networks enhances the granularity and reliability of DT environments, particularly in smart infrastructure systems.

The emergence of SC DTs is largely driven by the need to address complex urban challenges, such as traffic congestion, resource management and environmental sustainability. By simulating urban dynamics, DTs enable the exploration of different interventions before implementation, thereby minimising risks and optimising outcomes, as demonstrated by Campo et al. [5]. For instance, city officials can utilise these models to assess the impact of new traffic regulations or urban development projects, thereby making informed decisions that align with the city's strategic goals, as indicated by Dembski et al. [7]. Shamlitsky et al. [8] add to this by showing how DTs can support the effective management of traffic flows and improvements in transportation system performance. Furthermore, the ability to integrate real-time sensor data enhances the responsiveness of urban systems, allowing for proactive management of infrastructure and services, as illustrated by Rantanen et al. [9].

Integrating DT technology into the realm of eHealth within SCs represents a significant advancement in the management and delivery of healthcare services. A DT, as a virtual representation of a physical entity, allows for real-time monitoring, analysis and optimisation of health-related processes and systems as outlined by Kamel Boulos and Zhang [10]. Drawing on their work, DTs leverage data from various sources, including wearable health devices, eHealth records and environmental sensors, to create dynamic models that reflect the health status of individuals and populations. Additionally, Botín-Sanabria et al. [4] highlight the potential of DTs in public health applications, including preventive care and emergency response coordination. These findings facilitate a comprehensive understanding of urban health dynamics.

As urban environments continue to evolve in complexity, implementing digital health twins becomes crucial for enhancing the health and well-being of our communities. D'Hauwers et al. [11] stress the importance of using health-centric digital twins to simulate public health scenarios, optimise resource allocation and test intervention strategies before they are implemented. Moreover, the use of DTs in eHealth can significantly improve patient outcomes through personalised medicine. By creating individualised digital replicas of patients, healthcare providers can tailor treatments based on real-time data, leading to more effective interventions, as discussed by Botín-Sanabria et al. [4]. This capability is particularly relevant in managing chronic diseases, where continuous monitoring and timely adjustments to treatment plans are crucial, as emphasised by Kamel Boulos and Zhang [10]. Furthermore, the integration of artificial intelligence (AI) with DT technology can enhance predictive analytics, allowing for early detection of health issues and proactive management of patient care, as noted by Vempati and Mahajan [12].

On the other hand, by simulating various health scenarios and interventions, city officials can assess the impact of urban policies on public health outcomes as demonstrated by Hämäläinen [13]. For example, DTs can be used to model the effects of environmental changes, such as air quality improvements or the introduction of green spaces, on community health, as discussed by Fedwa et al. [14]. This capability is further elaborated by d’Hollosy et al. [15], particularly in the context of addressing public health challenges, such as the spread of infectious diseases or the management of chronic conditions, where timely and data-driven interventions are essential.

From the above, it is evident that SC DT can have a significant influence on the public health of city residents either by simulating the spread of diseases based on real time data, by continuously monitoring residents with health issues to create their digital replicas or by continuously modelling the impact of urban policies on public health. However, the implementation of an SC DT architecture, in general, and its combination with eHealth, in particular, comes with challenges. One of the foremost challenges is data management. In SCs, vast amounts of data are collected from numerous sources, including sensors, IoT devices, and environmental monitoring systems. These data, although abundant, are often fragmented and stored in isolated systems, making it difficult to create a unified real-time representation or digital replicas of corresponding entities, as highlighted by Allam and Jones [16]. This, in turn, reduces the ability of cities to simulate the spread of diseases faithfully, limits the effectiveness of monitoring the impact of the environment on unhealthy residents and hinders the capability of predictive analytics in the proactive management of residents' healthcare.

The other challenge is that issues related to data privacy, security and the interoperability of health information systems pose significant barriers to the widespread adoption of DT technology as explained by Ivanov et al. [2]. Consequently, the security and privacy of SC data in general and eHealth data in particular have been the focus of many researchers recently as highlighted by Ahmadi-Assalemi et al. [17] and Rafik et al. [18].

Thus, this paper introduces an SC DT architecture with a realisation of this architecture in the form of an open-source software tool [1] that can interface with real IoT devices or simulate these devices by generating correlated data from any available dataset. These real or simulated data are streamed through the tool using the message queueing telemetry transport (MQTT) protocol to a big data platform that categorises these data into relevant topics, stores it and makes it accessible to intelligent machine learning (ML) analytic models created by the user based on the use case and application demand. The tool is scalable in terms of the number of devices it can receive data from, either real hardware or simulated ones, and the number of employed ML models. The tool does not implement specific security functions since different proposed security and privacy schemes, such as those mentioned by Ahmadi-Assalemi et al. [17] and Rafik et al. [18], can be easily integrated into it due to its open-source nature as mandated by the SC municipality policies.

Moreover, the paper introduces three integrated case studies that demonstrate the proposed tool as a realisation of an SC DT providing a personalised public health advising system based on environmental factors, such as air quality level and health status of city residents.

The main contributions of this paper are threefold.

• It introduces a simple SC DT architecture that can be used for different SC use cases, including environment and eHealth-related ones, by accommodating IoT-sensed data and/or eHealth data from wearable devices of city residents in a big data analytics framework.

• It provides a realisation of the proposed architecture in the form of a scalable open-source software tool that can receive live data or simulate the generation of realistic data, stream it to a big data platform and analyse it using various configurable ML models.

• Different case studies are presented demonstrating the ability of the tool to employ AI to predict the impact of the environment on the health of SC residents with medical conditions.

The rest of the paper is organised as follows. Section 2 reviews the most relevant research works in the literature. Section 3 introduces an SC DT architecture fostering intelligent big data analytics. The structure of the software tool realising the introduced architecture is described in Section 4. Section 5 presents different case studies that demonstrate the usage of the proposed tool. Section 6 discusses the role of the proposed tool in supporting the digital transformation of cities into SCs, while also addressing potential limitations and possible mitigation strategies. Finally, Section 7 concludes the paper.

Related Works

The role of DTs extends beyond mere simulation as shown in several recent studies. Jeddoub et al. [19] discussed the gap between conceptual definitions of DTs and their current practical uses, with a focus on the difficulties in achieving effective data integration and system coordination. Wang et al. in ref. [20] identified the main challenges in applying DTs for SC management, emphasising the importance of real-time data integration. Attaran and Celik [21] examined the use of DTs in various industries and pointed out ongoing challenges, including data security, interoperability between systems and scalability concerns.

Indeed, DTs can play a vital role in enhancing urban resilience and sustainability. As cities face increasing pressures from climate change, population growth and technological advancements, DTs provide a framework for understanding and mitigating risks associated with urbanisation.

In urban planning and management, Ziehl et al. [22] evaluate how real-time digital monitoring can support adaptive planning in rapidly changing urban environments. They facilitate data-driven policy-making by offering comprehensive insights into urban environments, enabling stakeholders to identify vulnerabilities and develop effective countermeasures. Y. Huang et al. [23] argue that leveraging DT data can significantly enhance urban management by enabling real-time remote control of physical assets. Schrotter and Hürzeler [24] showcased the DT of Zurich as an example of how 3D spatial data can support urban planning and increase public involvement in city development. Allam and Jones [16] emphasised the future role of DTs in sustainable urban development, particularly when combined with next-generation technologies such as 6G and immersive environments. Hämäläinen [13] discusses the early adoption of DT technology in Helsinki, highlighting its potential to enhance urban development through improved data-driven decision-making. Moreover, the sociotechnical perspective on DTs underscores the importance of integrating social dimensions into urban planning, ensuring that the needs and behaviours of citizens are reflected in the decision-making process, as noted by Ravid and Rabinovitch [25]. Nochta et al. [26] reinforce this perspective by proposing policy frameworks that embed DTs within participatory urban governance models.

In addition to their application in urban planning and management, DTs are instrumental in disaster risk management. Shaharuddin et al. [27] argue that the ability to simulate various disaster scenarios allows cities to prepare and respond more effectively to emergencies. Ford and Wolf [28] propose a conceptual model for using SC DT in disaster management, illustrating how integrated sensing and simulation can improve community resilience. Ariyachandra and Wedawatta [29] highlight the importance of disaster preparedness using DTs, focusing on their role in emergency simulations. Astarita et al. [30] extend this by discussing multi-modal sensor integration within DTs, enabling better situational awareness and emergency response planning. This capability underscores the potential of DTs not only to improve operational efficiency but also to enhance the overall quality of life for urban residents.

In transportation, DTs also play a crucial role in optimising traffic management and enhancing mobility systems. Vempati and Mahajan [12] provide evidence for this by demonstrating the utility of DT-based monitoring in transportation and air quality assessment. Yeon et al. [31] introduce DTUMOS, a DT framework designed for urban mobility operating systems, which facilitates the testing of various mobility algorithms and policies.

Thus, from the above, it is demonstrated that DTs can be employed across various domains within smart cities, including urban planning, environment protection, infrastructure management, transportation systems, situational awareness and public health, as also highlighted by Allam and Jones [16]. This, indeed, requires the engagement of new technologies to realise the goals of SCs. Recent research by Szpilko, Fernando et al. [32] explores the strategic role of technologies, such as DTs, AI and IoT, in energy optimisation and climate-conscious urban planning. Their findings advocate for smart grid development and real-time data integration for effective urban energy management. Furthermore, Kuru [33] proposes the MetaOmniCity framework, a novel concept that fuses DTs and metaverse technologies, to create immersive 3D urban environments. The study explores how such frameworks can enhance citizen interaction, remote service delivery and real-time visualisation of city functions. This broader vision highlights the transformative potential of SC DTs in enabling spatially and temporally independent urban services.

The benefits of employing DTs in eHealth are manifold. They have been increasingly applied in various eHealth contexts, including personalised healthcare, chronic disease management and health monitoring systems, since they enable healthcare providers to simulate various health scenarios, assess the impact of interventions and optimise resource allocation, as explained by Ven et al. [34]. P. Huang et al. [35] discuss the potential of DTs to provide personalised healthcare services, emphasising the need for ethical considerations in their deployment. Ferko et al. [36] emphasise the importance of architectural solutions for DTs, which can evaluate ‘what-if’ scenarios using intelligent algorithms. d’Hollosy et al. [15] present an interoperable eHealth reference architecture that facilitates the integration of DTs into primary care using a service-oriented approach and a communication bus to connect distributed applications. Luazuaroiu et al. [37] emphasise the potential of optimising hospital operations and patient care during the COVID-19 pandemic with deep ML algorithms. Their work highlights the critical role of AI-based decision systems in predicting ICU admissions and mortality risk as well as supporting real-time clinical decisions based on vital signs and medical histories. These findings underscore the value of integrating intelligent analytics into DT-based healthcare systems.

The integration of DTs with other technologies, such as IoT and AI, is critical for enhancing their effectiveness in eHealth applications. Qian et al. [38] discuss the role of IoT in connecting smart devices to collect and analyse health data, enabling effective monitoring for healthcare purposes. Szpilko, Naharro et al. [39] showed that AI applications in smart cities span several critical domains, including mobility, safety, energy and public health. Their review reveals gaps in AI-driven systems concerning data security, integration with IoT and standardisation across sectors. This supports the need for holistic DT frameworks that embed AI for multi-domain urban decision-making while maintaining ethical and secure data practices.

In short, some research works in the literature discuss the capabilities of DTs to enable SCs to improve city governance and citizen welfare by enhancing urban resilience via planning, management and sustainability. Other works demonstrate that DTs are also envisioned to support community resilience through disaster management and mobility enhancement as seen in the conceptual model proposed by Ford and Wolf [28] or the operational framework illustrated by Yeon et al. [31]. The benefits of employing DTs in healthcare have been highlighted by other researchers as mentioned above. However, a few research works have provided insights on how integrating IoT and AI in SC DTs can enable effective eHealth applications as demonstrated by Qian et al. [38] and Szpilko, Naharro et al. [39].

To the best of our knowledge, no other research work in the literature offers a tool that provides a realisation of SC DT architecture, thereby fostering the integration of IoT, big data and ML, and allowing for the modelling of the impact of environmental changes on community health.

Digital Twin Proposed Layered Architecture

Our study emphasises using DT technologies within smart cities while supporting eHealth for city residents, demonstrating its potential to transform these domains via data-driven decision-making and predictive analytics. In the sequel, we introduce a simplified DT architecture for SC services including eHealth.

As illustrated in Figure 1, the proposed architecture consists of multiple layers, each contributing to the overall functionality of the DT system. These layers include the data emulation layer, data exchange layer, data ingestion layer, data intelligence layer, decision support layer and client interaction layer. Each layer plays an essential role, working together to ensure the smooth operation of the system.

[IMAGE OMITTED. SEE PDF]

In the figure, we also see a depiction of an SC with sensors, highlighting the extensive use of sensors that represent the cyber–physical world, where everything revolves around data. SCs rely on a broad spectrum of IoT sensors, including those for detecting motion, tracking locations, monitoring air quality, temperature, wind speed, humidity and many more. In addition, there are wearable devices for city residents that can be in different forms, such as smartwatches and portable activity/health monitoring devices, carrying various sensors, such as heartbeat, respiratory rate and stress level, to name a few. The data generated by these sensors propagate through the architecture, playing a crucial role in how the system functions. All these sensors hold a massive amount of valuable information. Thus, the architecture is designed to handle and process this amount of data efficiently. Additionally, the architecture allows for flexibility and scalability, making it easier to adapt to different applications and requirements. The specific functions of the layers are discussed below.

Data Emulation Layer

The data emulation layer is the first layer in the DT architecture and plays a critical role in generating data when real-time sensor data from physical sensors is not yet available for the required scenarios. This layer should contain emulated sensors that act, such as real sensors, providing the ability to be configured to generate synthetic correlated data in real time based on historical datasets. It is mainly used to simulate the data generation for intended use cases, especially when the available data are insufficient for intelligent analytics, helping maintain predictive and classification capabilities of the DT system with acceptable accuracy for decision-making functions.

Data Exchange Layer

This layer will stream the data from the cyber–physical world, or emulated sensors, in real time. By using standard data communication protocols, such as message queueing telemetry transport (MQTT), as described by Kantawong et al. [40], or the constrained application protocol (CoAP) for large-scale IoT deployments, as mentioned by Bormann et al. [41], this layer ensures that data move efficiently and securely throughout the system. In smart cities, for instance, data collected from sensors such as those monitoring temperature, traffic, or pollution levels need to be transmitted reliably. The data exchange layer also ensures this happens without delay, providing real-time transfer to critical information, such as eHealth data. Thus, it shall support communicating with IoT devices, wearables or portable devices via different wireless technologies. In addition, this layer is responsible for preserving the main security and privacy functions related to the received data, such as confidentiality, integrity and authentication of these data as well as the authentication of the hardware devices that send it.

Data Ingestion Layer

Once data have been collected and transferred to the DT system, the data exchange layer passes it to the data ingestion layer. This layer is responsible for dealing with data streamed in real time. It stores these data, organises it according to its topics or categories and prepares it to be analysed effectively. It ensures that the incoming data are structured, valid and ready for further processing, thereby maintaining the quality and reliability of the data being used. The datasets stored by this layer can also be used to support the data emulation layer in simulating new scenarios or use cases that require information from these data.

Data Intelligence Layer

The data intelligence layer is responsible for analysing and processing data contained in the data ingestion layer to uncover meaningful insights. This layer uses techniques, such as ML and data mining, to build and optimise models, making them available for further evaluation. It is also tasked with training and testing these models to ensure accuracy and performance. Additionally, this layer supports anomaly detection by identifying irregularities or unexpected patterns in the data, which is crucial for detecting potential issues such as faulty sensors or unusual system behaviour. Through continuous learning, the models are regularly updated by the data ingestion layer with new data, allowing the system to adapt and improve over time.

Decision Support Layer

After the data have been processed, this layer provides guidance to users, such as city municipality officials or healthcare professionals, on how to respond to various situations. It leverages the ML models from the data intelligence layer to make informed decisions based on the processed data. For example, in an eHealth-related scenario, the system may alert doctors to a patient's deteriorating condition, or in another scenario, it could provide forecasts on the high-energy consumption of certain city areas or buildings based on weather conditions or other factors. Therefore, this layer translates data-driven insights into practical actions that enhance decision-making and overall system efficiency.

Client Interaction Layer

The client interaction layer is the top layer, where the DT system client or user engages directly with it. This layer serves as the interface for users, including SC officials, healthcare professionals and city residents, to control, monitor or interact with the system, respectively. It shall offer a portable user-friendly interface that is intuitive for nontechnical users, allowing them to interact with the system effortlessly. Users can visualise information based on their specific needs. For instance, it can offer a forecasting service regarding the anticipated status of certain public parking areas for city residents, recommending the best place to park their cars in the near future. Additionally, the layer includes an alert and notification system that sends timely alerts for any critical changes or issues, enabling immediate action when necessary. For example, in an SC, users might receive alerts on energy consumption spikes or traffic congestion. Similarly, in an eHealth application, doctors could be notified about sudden changes in a patient's condition. This layer is crucial for ensuring decision-makers can act quickly and effectively based on real-time data. Thus, this layer is responsible for providing confidentiality, integrity and authentication for the exchanged data. In addition, it should support authentication and access control for all clients dealing with the DT system. Furthermore, this layer can also interact with client devices in a cyber–physical system (CPS) if these devices are required to be automatically controlled based on the decisions made by the decision support layer, which relies on analytics supported by real or simulated data.

The layered structure of the DT architecture enables the seamless integration of different technologies, providing the flexibility necessary to manage complex systems, such as smart cities. The layering ensures that each function, from data collection to decision-making, is handled efficiently and can be adapted to suit a wide range of use cases.

A Software Realisation for the Layered DT Architecture for Smart Cities

In order to realise the aforementioned simplified layered architecture, we propose a modular open-source tool called the digital twin realisation for smart city (DTRSC) tool. The tool realises most of the proposed functionalities of each layer of the layered architecture. For seamless integration and easy maintenance, each layer of the proposed DT architecture is implemented as an independent module using Python code. The tool features a command-line interface (CLI) as the primary means of interaction, allowing users to configure, monitor and engage with the DT model in a scalable, portable and computing-resources-efficient manner. It enables users to interact directly with SC services, including eHealth applications, making more effective use of real-time data.

The adaptable nature of the DTRSC tool ensures that all components are tightly integrated with the overall DT framework, allowing for smooth transitions from data generation to real-time analysis and predictive modelling. This approach not only optimises dynamic SC functions, such as traffic management and environment protection, but also extends its capabilities to healthcare applications, including patient status monitoring and predictive diagnostics. DTRSC is built with open-source technologies and a modular architecture that ensures robustness while allowing local authorities and developers to adapt and expand the tool according to city-specific requirements without altering the core structure.

The DTRSC Structure

The DTRSC tool is designed as an open-source platform, providing broader accessibility and customisation for various applications. The main components and their interactions are illustrated in Figure 2, demonstrating how each module integrates into the DT framework to deliver seamless data generation, real-time processing and predictive analytics. In the sequel, the functionality of each tool module will be presented.

[IMAGE OMITTED. SEE PDF]

Data Emulation Module

The data emulation module aligns with the data emulation layer of the proposed DT architecture, containing emulated devices that replicate real-world IoT nodes. Each device can contain multiple sensors of different types. The emulated sensors are developed using pre-trained ML models created from realistic datasets. When real-time sensor data are unavailable or difficult to obtain, the system leverages these models to generate a time series of synthetic correlated data in real time. The generated data closely mirror real-world conditions, accurately imitating the performance of physical sensors. Similar to physical IoT devices, the emulated devices communicate with the rest of the tool (the data transfer module) via the MQTT protocol. Therefore, every emulated device should be assigned an IP address. These emulated devices can be created individually or in groups. This allows the tool to support the scale required for smart cities in terms of the number of devices.

A configurable data generation profile controls the data generation process, allowing users to define the type of emulated sensors they need to attach to any virtual (emulated) device. The list of available sensors can be expanded by integrating more emulated sensors, as the DTRSC tool is open-source. Moreover, the time profile for the data generation is configurable. It can be periodic or random data generation, depending on the user's preference, based on the type of data to be generated. Additionally, the emulation time can be customised, providing flexibility for different scenarios.

Data Transfer Module

The data transfer module corresponds to the data exchange layer of the DT architecture and is responsible for the efficient streaming and exchange of data. Once data are generated by real and/or emulated sensors, it is encapsulated into MQTT network packets, which are then transmitted to an MQTT broker. The broker manages message routing between the sensors (real or emulated) and the rest of the tool. In this setup, hardware devices and emulated sensors operate as MQTT clients, whereas the MQTT broker ensures smooth data flow within the system. To function as clients, hardware/emulated devices must be assigned IP addresses. For moving devices, such as eHealth gadgets, a mobile IP scheme is assumed to be in place.

To enhance the data exchange process, the DTRSC uses PyShark [42] to capture and analyse network traffic, providing a unified interface to the big data handling module regardless of the data source (real hardware or emulated sensors). This makes data generated by sensors, whether real or emulated, flow seamlessly through the system, allowing for continuous real-time analysis and predictive modelling across various applications. Although MQTT is the primary protocol used for communication, the DTRSC is flexible enough to support alternative data protocols if needed.

Big Data Handling Module

The big data handling module is a critical component that manages the continuous data flow within the DT system. The properties of the data ingestion layer are implemented here by leveraging Kafka's [43] robust streaming capabilities. This module captures and processes real-time data from physical devices or emulated sensors. The data captured by PyShark are then stored in pre-assigned Kafka topics, categorised by device ID or IP address and sensor type. These topics shall be created based on the use case or the DT service offered by SC officials. The mapping between device IDs or IP addresses, locations and their owners is inaccessible to the DTRSC tool and its users. They should be kept secret in confidential records to maintain the anonymity of these devices for any SC DT service, including eHealth.

This module can store data for further analysis or send the information to the data emulation module to simulate devices. Kafka's configuration is optimised for performance, with features, such as message compression, idempotency and batch processing, to enhance data throughput and reliability. This ensures that the data pipeline can efficiently handle large volumes of information, which is crucial for maintaining real-time analytics capabilities in different SC DT services.

Data Processing and Analytics Module

The data processing and analytics module forms the analytical backbone of the DTRSC, representing the data intelligence and client interaction layers of the DT architecture. It processes real-time data streams and historical datasets to extract actionable insights and generate predictions. Users can select specific ML models through the CLI, such as long short-term memory (LSTM), linear regression, random forest, gradient boosting or neural networks, to perform regression and classification tasks, but it can also be extended to include additional models. This module's flexibility allows it to cater to a wide range of analytical needs, supporting complex data-driven decision-making.

During the data processing phase, the system applies data pre-processing techniques to ensure the information is ready for ML analysis, addressing aspects such as missing values, feature scaling and data encoding. Once the data are refined, the models generate predictions, delivering insights that guide strategic actions for city authorities or tool users.

The CLI-based interface empowers authorities or tool users to interact directly with the system, allowing them to configure the ML models, make enquiries and receive immediate responses even with limited programming knowledge.

The DTRSC Work Flow

The DTRSC tool developed in this study leverages a CLI designed to manage and configure the entire workflow from data generation to analytics. The DTRSC acts as an intuitive platform that allows users or relevant authorities to set up and control DT applications across various domains, such as eHealth, air quality monitoring and beyond. This section details the workflow experienced while using the DTRSC, illustrating the interaction process at each stage.

Upon launching the CLI, users are prompted to either start a new project, load an existing project or exit the application. Selecting ‘Start a new project’ initiates a new project configuration as shown in Figure 3. After creating the project, the DTRSC prompts the user to choose between hardware-generated or synthetic data. In this figure, the user selects the option to proceed with synthetic data.

[IMAGE OMITTED. SEE PDF]

The data emulation module is responsible for configuring the network of IoT devices or sensor nodes. The user first provides a use case such as eHealth in Figure 4. This is labelled as ‘Case 1.’ The user then chooses how to add devices based on the application requirements, with options to add a group or a single device. Figure 4 illustrates the process of adding a group of identical devices. Here, the user specifies the number of devices, selects sensors for the first device and the remaining devices and the MQTT broker will be automatically configured as shown in Figure 5.

[IMAGE OMITTED. SEE PDF]

[IMAGE OMITTED. SEE PDF]

If the user selects the single device option, it can be configured by naming the device, assigning an IP address for the MQTT client and choosing the sensors to be used as shown in Figure 6. After completing the network configuration, the system sets up the MQTT broker, which manages communication between devices in the data exchange layer. If devices are added individually, the user can view a configuration summary of each device. The user is also prompted to choose a time profile (periodic or random) for data generation, allowing flexibility in emulating real-time sensor device behaviour.

[IMAGE OMITTED. SEE PDF]

Once the sensor devices are configured, the user is prompted to initiate data streaming. The big data handling module then begins streaming the generated sensor or IoT data into a Kafka-based big data platform by connecting to the Kafka broker and creating Kafka topics to categorise data by sensor type as shown in Figure 7. For instance, topics, such as D1_Temperature and D1_HeartRate, are created to stream temperature and heart rate data in real time. In this example, we illustrate five topics due to screen limitations, which restrict capturing additional topics in a single image. Kafka's producer configurations ensure that data are efficiently streamed to the correct topics while consumers subscribe to these topics to read data as it is generated. Users can monitor the data stream and choose to view the streaming data in real time through Kafka as shown in Figure 7.

[IMAGE OMITTED. SEE PDF]

After the data have been streamed into Kafka, the next step is to analyse it using the data processing and analytics module. The DTRSC tool allows users to load an existing dataset or generate a new real-time dataset for analysis as shown in Figure 8. For real-time dataset generation, users specify the duration for which data will be captured from the real time streamed data. They can then select input variables from the available Kafka topics. For example, a ML model may use selected input features from this list to predict the target variable(s). Here, in Figure 8, respiratory rate is chosen as the target (output) variable.

[IMAGE OMITTED. SEE PDF]

Once the input and output variables are selected, the system allows the user to choose a ML algorithm from a list of available options, including LSTM (long short-term memory), linear regression, gradient boosting, random forest and neural networks, as shown in Figure 9. The tool can be easily extended to include other models and their respective parameters. The DTRSC tool supports the creation of both regression and classification models, and in this example, regression analysis is performed. The user selects a neural network model, and the tool prompts them to configure it by specifying parameters such as the number of layers, neurons per layer and the optimisation algorithm. For instance, a neural network can be configured to include five layers and 64 neurons per layer and to use the Adam optimisation algorithm. The model is trained using the dataset generated from the streaming data, with 80% of the data used for training and 20% for testing. The training process involves fitting the model to the data, and the system provides feedback on the performance of the trained model.

[IMAGE OMITTED. SEE PDF]

Once the model has been trained, users can obtain the output values by providing input values for the selected features. The DTRSC tool then uses the trained model to predict the output variable and display the result to the user.

At this stage, users can choose to continue using the current model, create a new one or exit the application. This flexibility allows the DTRSC tool to support iterative testing and model refinement. After the analysis is completed, the results generated by the ML model are communicated back to the user of the DTRSC tool. Decisions derived from the predictive analysis can be shared with intended stakeholders, such as healthcare providers or city authorities, enabling real-time interventions. Additionally, users with little background in ML and computer programming can leverage this tool to perform predictive analysis effectively.

Case Studies

This section highlights how the DTRSC tool's design provides authorities and users with responsive real-time functionality for managing and adapting to dynamic conditions in an SC. By leveraging the proposed DT layered architecture, the DTRSC tool can support a multitude of applications that aim to improve citizens' Quality Of Life (QoL). Here, we demonstrate the ability of the tool to handle use cases analysed using different ML models and to combine their results in another use case built on the same data topics streamed to the big data platform.

The first use case focuses on forecasting air quality in smart cities, offering valuable insights for public health management and individual recreation. The second use case involves predicting an individual's respiratory condition based on vital signs collected from wearable devices, facilitating personalised health monitoring. The third use case combines the air quality forecast with the individual's respiratory condition to classify each situation as an ‘alert’ or ‘no alert.’ These real-time alerts are generated by analysing the individuals' vital signs alongside the air quality index (AQI) to assess whether the anticipated outdoor conditions are safe for them. To evaluate the system's effectiveness, we emulated the generation of real-time data from individuals' datasets and assessed the tool's accuracy in generating these alerts.

It is worth noting that these use cases represent examples or case studies. The tool design based on the proposed DT architecture can be adapted to implement various other applications.

Case Study 1: AQI Forecasting in Smart Cities

As highlighted by the World Health Organization (WHO), outdoor air pollution remains a significant global public health concern, contributing to different respiratory diseases as highlighted by the World Health Organization [44]. Given the impact of air pollution on health, particularly in urban areas, we chose air quality forecasting as the first use case for our study as emphasised by Marino et al. [45]. This use case demonstrates how the DTRSC tool can be used to monitor and forecast the AQI in real time, helping city authorities and individuals take preventive measures when air quality declines.

In this case study, we utilised a time-series dataset from Zheng et al. [46] containing real-time measurements of key air pollutants, including particulate matter (PM2.5 and PM10), ozone (O3), nitrogen dioxide (NO2), sulphur dioxide (SO2) and carbon monoxide (CO), as well as meteorological variables such as temperature, humidity and wind speed. These variables were used as input features for forecasting the AQI, which serves as the output variable. Using the emulation module of the DTRSC tool, we generated real-time data from the dataset. We augmented the dataset with the AQI values, which are calculated using Equation (1) for each pollutant, as outlined by the U.S. Environmental Protection Agency [47], 1 $$AQ{I}_{p}=\frac{\left({I}_{\text{high}}-{I}_{\text{low}}\right)}{\left({C}_{\text{high}}-{C}_{\text{low}}\right)}\times \left({C}_{p}-{C}_{\text{low}}\right)+{I}_{\text{low}}$$ where $$AQ{I}_{p}$$ represents the AQI for pollutant $$p$$ and $${C}_{p}$$ is the measured concentration of pollutant $$p$$. The terms $${C}_{\text{high}}$$ and $${C}_{\text{low}}$$ denote the breakpoint concentrations for the upper and lower bounds of the pollutant's concentration range that contains $${C}_{p}$$. Similarly, $${I}_{\text{high}}$$ and $${I}_{\text{low}}$$ represent the upper and lower bounds of the AQI scale for pollutant $$p$$ at the concentration levels $${C}_{\text{high}}$$ and $${C}_{\text{low}}$$, respectively. This formula ensures AQI calculations align with EPA guidelines by basing values on pollutant concentrations [47].

The AQI for each pollutant in a given row of data is calculated individually using the formula above, whereas the overall AQI for that row is determined by selecting the maximum AQI value among all pollutants. This approach follows standard practices, where overall air quality is determined by the pollutant with the highest individual AQI value at any given time.

Indeed, the AQI categorises air quality into different severity levels based on the concentrations of pollutants. Table 1 outlines the breakpoints for the AQI, following the Environmental Protection Agency (EPA) guidelines [47].

TABLE 1 Breakpoints for AQI calculation for ozone ($${\mathrm{O}}_{3}$$), $${\text{PM}}_{2.5}$$, $${\text{PM}}_{10}$$, $${\text{NO}}_{2}$$, CO and $${\text{SO}}_{2}$$.

AQI category	O3 (ppm)	$${\text{PM}}_{2.5}$$ (μg/m3)	$${\text{PM}}_{10}$$ (μg/m3)	$${\text{NO}}_{2}$$ (ppb)	CO (ppm)	$${\text{SO}}_{2}$$ (ppb)
Good (0–50)	0.000–0.054	0.0–12.0	0–54	0–53	0.0–4.4	0–35
Moderate (51–100)	0.055–0.070	12.1–35.4	55–154	54–100	4.5–9.4	36–75
Unhealthy for sensitive groups (101–150)	0.071–0.085	35.5–55.4	155–254	101–360	9.5–12.4	76–185
Unhealthy (151–200)	0.086–0.105	55.5–150.4	255–354	361–649	12.5–15.4	186–304
Very unhealthy (201–300)	0.106–0.200	150.5–250.4	355–424	650–1249	15.5–30.4	305–604
Hazardous (301–500)	—	250.5–500.4	425–604	1250–2049	30.5–50.4	605–1004

To forecast the AQI levels, we employed the LSTM model, which is known to have a unique architecture with input, forget and output gates. These gates control what information is added, retained or discarded, allowing the model to focus on meaningful patterns over longer time spans. This capability makes the LSTM model particularly effective for handling the complexities of AQI data, including seasonal trends, sudden fluctuations and long-term variations, as shown by Yadav et al. [48]. To evaluate the effectiveness of the LSTM predictions, we applied several performance metrics implemented in the tool, which are discussed in detail in the sequel, following the works of Dubey et al. [49], Tatachar [50] and Chicco et al. [51].

Performance Metrics

Mean Absolute Error (MAE)

The mean absolute error (MAE) measures the average magnitude of the errors between the actual values and the predicted values without considering the error sign. It is defined as follows: 2 $$\text{MAE}=\frac{1}{m}\sum\limits _{j=1}^{m}\left\vert {a}_{j}-{\hat{p}}_{j}\right\vert $$ where $$m$$ represents the total number of predictions, $${a}_{j}$$ is the actual value and $${\hat{p}}_{j}$$ is the predicted value [49, 50]. This metric represents the prediction error by measuring the average absolute difference between actual and predicted values.

Root Mean Squared Error (RMSE)

The root mean squared error (RMSE) is the square root of the mean squared error (MSE), providing a measure of error in the same units as the predicted variable. It can be obtained from 3 $$\text{RMSE}=\sqrt{\frac{1}{m}\sum\limits _{j=1}^{m}{({a}_{j}-{\hat{p}}_{j})}^{2}}$$ where $$m$$ represents the total number of predictions, $${a}_{j}$$ is the actual value and $${\hat{p}}_{j}$$ is the predicted value. This metric is particularly useful for understanding the standard deviation of prediction errors and is sensitive to larger errors [49, 50].

Mean Absolute Percentage Error (MAPE)

The mean absolute percentage error (MAPE) represents the prediction error as a percentage, providing a relative measure of the error. It can be calculated from 4 $$\text{MAPE}=\frac{1}{m}\sum\limits _{j=1}^{m}\left\vert \frac{{a}_{j}-{\hat{p}}_{j}}{{a}_{j}}\right\vert \ast 100$$ where $$m$$ represents the total number of predictions, $${a}_{j}$$ is the actual value and $${\hat{p}}_{j}$$ is the predicted value. MAPE is often used to compare prediction performance across models, as it indicates the average percentage error, making it useful for evaluating relative accuracy [49–51].

R-Squared $$\left({R}^{\mathbf{2}}\right)$$

The $${R}^{2}$$ (coefficient of determination) measures the proportion of variance in the dependent variable that can be explained by the independent variables. It is expressed as follows: 5 $${R}^{2}=1-\frac{{\sum }_{j=1}^{m}{({a}_{j}-{\hat{p}}_{j})}^{2}}{{\sum }_{j=1}^{m}{\left({a}_{j}-\overline{a}\right)}^{2}}$$ where $${a}_{j}$$ is the actual value, $${\hat{p}}_{j}$$ is the predicted value and $$\overline{a}$$ is the mean of the actual values. The $${R}^{2}$$ value ranges from 0 to 1, with higher values indicating better model performance, as it shows how well the model explains the variability of the target variable [49–51].

Table 2 provides the results for the performance metrics of these models,

TABLE 2 Performance metrics of the LSTM model for AQI forecasting.

Model	MAE	RMSE	$${\boldsymbol{R}}^{\boldsymbol{2}}$$	MAPE
LSTM	0.7043	0.8541	0.88	0.2351

The Case Study 1 model results in Table 2 demonstrate the performance of the LSTM model for forecasting AQI with good accuracy, whereas processing data directly without the use of outlier handling and feature engineering techniques. Note that the LSTM model is used as an example, where a relatively short-term forecast is assumed for the purpose of public health awareness for outdoor activities. Other forecasting models can be implemented in the tool to provide long-term forecasting with sufficient accuracy.

The case study can be implemented with the aid of sensor nodes (IoT nodes) equipped with different types of air pollutant sensors and meteorological sensors (e.g., temperature, humidity and wind speed). These sensor nodes can be installed in various locations across the city, especially in areas where pollution is more likely to occur, such as near industrial zones. These IoT nodes can be connected to the SC municipality via a secure city-wide Wi-Fi network, a typical service in many SC initiatives, as discussed by Zanella et al. [52]. The DTRSC tool is intended to operate in a data centre managed by the city municipality. The tool's data transfer module will handle data reception and streaming to the big data handling module, which, in turn, will be accessed by the data processing and analytics module for analysis and informed decision-making. The tool's user interface can be utilised by SC officials for AQI forecasting, which indeed helps them with state and situation awareness (SSA) regarding the spatiotemporal forecasting of city pollution levels. This information can be made available to city residents by developing a suitable web portal for the tool's current CLI.

The current results highlight the tool's capability to process SC sensor data, showcasing its adaptability in handling unrefined data inputs. By leveraging historical data, the DT realised by the tool can provide reliable AQI forecasts, which are crucial for anticipating future air quality trends and supporting public health and environmental management efforts.

Case Study 2: Monitoring Respiratory Conditions Using Wearable Devices

Vicente et al. [53] demonstrate that wearable devices offer a convenient means of continuously monitoring vital signs in real-time, such as respiratory rate, which can be indicative of an individual's overall health status. With the increasing prevalence of smartwatches, fitness trackers and smart rings, it is now possible to collect real-time data on respiratory rate, heart rate, blood oxygen levels and other physiological parameters as indicated by Jeong et al. [54]. The accessibility and ubiquity of these devices have made continuous health monitoring more practical and widespread, empowering individuals to take preventive steps in managing their health in an SC setting where the city residents are always connected, as noted by Kamel Boulos and Zhang [10].

In this eHealth-related case study, the proposed tool enables the early detection of potential health risks, such as those related to lung function, as demonstrated by Liaqat et al. [55]. We utilised data collected from daily wearable devices provided by Reddy [56] to predict respiratory conditions based on an individual's respiratory rate. It is a vital sign that directly reflects the state of an individual's respiratory system. Abnormal respiratory rates can indicate various health issues, such as asthma, chronic obstructive pulmonary disease (COPD) or respiratory infections. If respiratory rate data are unavailable, it can be approximated from heart rate using the method outlined by Garmin [57] by using, 6 $$\text{Respiratory}\,\text{Rate}\approx \frac{\text{Heart}\,\text{Rate}}{4}.$$

In this case study, respiratory rates are classified into general risk categories based on the risk level for the individual's health, as described by Ra et al. [58], as shown in Table 3.

TABLE 3 Classification of respiratory rate by risk level.

Risk level	Respiratory rate (breaths per minute)
Low risk (normal)	12–20
Moderate risk (elevated)	$${ >} $$ 20
High risk (below normal)	$${< } $$ 12
Critical (severe abnormalities)	Variable (with symptoms)

For the predictive analysis of this case study, we selected skin temperature, heart rate, respiratory rate and oxygen saturation as input features. We augmented the dataset with the individual's respiratory condition based on the risk level defined in Table 3. The classification models in DTRSC were applied for this purpose, showcasing the tool's capability to handle classification tasks effectively. The available classification models include logistic regression (LR), random forest (RF), gradient boosting (GB) and artificial neural networks (ANNs), with the flexibility to integrate additional models as needed based on application requirements. The performance of these classification models for the used dataset was evaluated using several classification metrics available in the proposed tool. The key metrics outlined by Vakili et al. [59] are summarised below.

Key Metrics

Accuracy

Accuracy is a measure of the proportion of correct predictions made by the model, combining both true positives $$(Tp)$$ and true negatives $$(Tn)$$ relative to the total cases. 7 $$\text{Accuracy}=\frac{Tp+Tn}{Tp+Tn+Fp+Fn}$$

In Equation (7), $$Tp$$ (true positives) represents cases where the model correctly identified positive instances and $$Tn$$ (true negatives) denotes correctly identified negative cases. Conversely, $$Fp$$ (false positives) are instances incorrectly classified as positive, whereas $$Fn$$ (false negatives) are cases wrongly identified as negative.

Precision

Precision indicates the proportion of positive predictions that were actually correct, showing the model's effectiveness in selecting relevant positive cases. 8 $$\text{Precision}=\frac{Tp}{Tp+Fp}$$

In Equation (8), $$Tp$$ refers to true positives, which are correctly predicted positive cases, whereas $$Fp$$ denotes false positives, representing cases mistakenly predicted as positive by the model.

Recall

Recall, also known as sensitivity, measures the proportion of actual positive cases that the model correctly identifies. 9 $$\text{Recall}=\frac{Tp}{Tp+Fn}$$

In Equation (9), $$Tp$$ represents true positives, or the positive cases correctly identified by the model, and $$Fn$$ represents false negatives, or cases where the model failed to identify actual positives.

F1 Score

The F1 score provides a balanced metric by taking the harmonic mean of precision and recall, especially useful in scenarios with imbalanced datasets. It is calculated as follows: 10 $$F1=2\times \frac{\text{Precision}\times \text{Recall}}{\text{Precision}+\text{Recall}}$$

AUC-ROC

The area under the receiver operating characteristic curve (AUC-ROC) assesses the model's effectiveness across various thresholds. The ROC curve is a plot of sensitivity $$Se$$ versus the false positive rate $$Fpr$$. 11 $$Se=\frac{Tp}{Tp+Fn}$$ 12 $$\quad Fpr=\frac{Fp}{Fp+Tn}$$

In Equation (11), $$Se$$ is equivalent to the true positive rate, whereas $$Fpr$$ in Equation (12) represents the proportion of negative instances incorrectly classified as positive. Here, $$Tp$$ denotes true positives, $$Fn$$ indicates false negatives, $$Fp$$ refers to false positives and $$Tn$$ represents true negatives. The AUC value provides a single measure summarising the area under the ROC curve, with higher values indicating better model performance.

The following table (Table 4) provides the results for the performance metrics of the second case study.

TABLE 4 Performance of different ML models for predicting respiratory conditions.

Model	Accuracy	Precision	Recall	F1 score	AUC-ROC
Logistic regression	0.9437	0.9435	0.9438	0.9436	0.9863
Random forest	0.9319	0.9312	0.9319	0.9301	0.9728
Gradient boosting	0.9475	0.9469	0.9475	0.9471	0.9839
Artificial neural networks	0.9363	0.9369	0.9363	0.9365	0.9817

As shown in Table 4, the DTRSC tool allows users to evaluate and compare ML models, such as LR, RF, GB and ANNs, each assessed across metrics such as accuracy, precision, recall, F1 score and AUC-ROC. To enhance these metrics, we appended additional emulated data provided by Bommela [60], achieving improved results with our classification models. This capability enables users to select the most suitable model based on performance, facilitating accurate predictions for conditions such as respiratory health issues. Table 4 shows GB achieves the highest accuracy (0.9475) and F1 score, followed closely by logistic regression. Random forest and artificial neural networks also perform well, with AUC-ROC values above 0.97, indicating strong classification capabilities across all models. With its intuitive, CLI-based interface, the DTRSC tool makes ML analytics accessible not only to technical users but also to city authorities, healthcare providers and others without extensive ML expertise.

This case study can be implemented via a smartphone application that can be provided by the SC's municipality to city residents via well-known smartphone software stores (e.g., Play Store). The application, after getting appropriate permission from the city resident user, can send the wearable eHealth user data to the DTRSC tool data transfer module via the SC secured Wi-Fi network to be streamed to the corresponding (pre-configured by city officials) big data topics through the big data handling module and made available to the data processing and analytics for analysis and decision-making. The user data can carry a particular ID or IP address based on the account the city user utilises to access the SC Wi-Fi. As mentioned before, the mapping between the user device ID or IP address will be kept confidential and not be revealed to the DTRSC tool admins or users to preserve the anonymity of city resident users. The analysis of eHealth wearable data can be vital for SSA, as it enables the issuance of personalised warnings to city residents at risk of respiratory complications. Moreover, it guides city officials toward identifying respiratory health risks that some residents may face, allowing these officials to relate these risks to the pollution status of different city locations.

Case Study 3: Outdoor Health Hazard Warning

This use case demonstrates the capability of the DTRSC tool to integrate the usage of the models of other use cases and their underlying datasets managed by the big data handling module into a new use case. Integrating air quality forecasting with wearable health data via the DTRSC tool advances the ability to create a robust health management system for SCs. By combining environmental and personal health data in real-time, this case study offers individuals tailored health alerts whenever their physiological metrics and surrounding environmental conditions reach critical levels.

Empirical studies by Marino et al. [45] and Götschi et al. [61] demonstrate a strong link between air quality and respiratory health, underscoring the importance of monitoring air quality alongside individual respiratory indicators. Thus, in this use case, as illustrated in Figure 10, we create a synthetic time series dataset that merges air quality data with wearable health metrics, emulating simultaneous data capture from both sources. This dataset provides a foundation for testing the model's ability to issue personalised health alerts that consider both environmental and individual health factors. Here, we consider the previously mentioned case studies, with their data being emulated and transferred via the data emulation and data transfer modules of the DTRSC tool and then streamed into the big data handling module. As seen in Figure 10, model topics are created for each case study and then aligned using timestamps generated by the data handling module. The DTRSC tool's classification model processes these combined topics effectively, generating timely warnings (alert) or no alert status based on the integrated data considering the respiratory rate and AQI index.

[IMAGE OMITTED. SEE PDF]

Table 5 presents the performance metrics of the third case study.

TABLE 5 Performance of different ML models for outdoors health warning use case.

Model	Accuracy	Precision	Recall	F1 score	AUC-ROC
Logistic regression	0.9725	0.9724	0.9725	0.9724	0.9963
Random forest	0.9725	0.9724	0.9725	0.9724	0.9936
Gradient boosting	0.9763	0.9761	0.9763	0.9761	0.9882
Artificial neural networks	0.9588	0.9592	0.9588	0.9589	0.9931

The results obtained from the DTRSC tool and presented in Table 5 show the performance of various ML models in generating accurate health warnings. Both the LR and random forest models achieve high accuracy, precision, recall and F1 scores of 0.9725, indicating strong and balanced performance in prediction reliability. Gradient boosting slightly outperforms these with accuracy and an F1 score of 0.9763, making it particularly effective in this application. Although the ANNs have a slightly lower accuracy of 0.9588, they still demonstrate strong performance, especially with an AUC-ROC of 0.9931, underscoring their ability to distinguish between alert and no-alert cases. Notably, all models achieve AUC-ROC values above 0.98, with logistic regression leading at 0.9963, indicating excellent classification capability across the models. These results suggest that each model is well-suited for generating timely and accurate health warnings, with GB providing a slight advantage in prediction accuracy and overall consistency.

The implementation of this case study mainly depends on the implementation of the other two, since it is conducted based on combining the big data topics of both. This use case can employ the service of the smartphone application mentioned in Case Study 2 to help SC residents plan for outdoor activities at a specific time in the future. The application can read the measured eHealth data from the user's wearable device and send it over the city's WiFi connection to the DTRSC tool, which can classify the user's respiratory health status and combine it with the AQI forecast at the time the user plans an outing. Personalised advice or recommendations based on the output of the DTRSC tool can be pushed via a portal to the smartphone application of city residents with hazardous respiratory health status as a part of the SSA offered by the SC municipality.

Discussion

The transformation of urban environments into SCs represents a profound shift in how city systems are designed, governed and experienced. An SC is not merely a collection of technologies but a holistic framework that integrates physical infrastructure, digital intelligence and human participation to improve QoL, promote sustainability and enable responsive governance as noted by Kuru [62]. As urban areas become increasingly complex and densely populated, there is a growing need for solutions that can operate in real-time and adapt to changing conditions. Technologies, such as the IoT, CPS, DTs and interoperable big data platforms, are central to this shift, enabling integrated sociotechnical systems that can respond to the evolving needs of urban populations. Among these, DTs play a pivotal role by combining data, models and human interaction to support informed decision-making in complex urban environments as highlighted by Mazzetto [63].

A DT creates a dynamic digital representation of physical objects, systems or processes, allowing for ongoing monitoring, simulation and optimisation. When applied at the city level, Urban Digital Twins (UDTs) offer a unified view of urban operations by combining data from diverse sources to support planning and operational decisions as discussed by Argota et al. [64]. UDTs are also becoming increasingly connected with metaverse applications, where virtual and augmented environments provide new ways to engage citizens and stakeholders in city development as noted by Kaya [33]. Indeed, digital twins, as virtual replicas of physical entities, when combined with metaverse technologies, such as virtual reality and augmented reality, can enhance urban planning and management as pointed out by Lv et al. [65]. When DTs are integrated with the metaverse, the information provided by the DTs can be replicated in the metaverse as a platform for citizens to engage with proposed changes in urban planning and policy as discussed by White et al. [66]. This synergy enables immersive experiences and improved decision-making in smart cities as discussed by Aloqaily et al. [67]. As Krylov suggests [68], this blending of physical and digital urban layers opens up new possibilities for more inclusive, responsive and forward-looking approaches to urban management.

Towards a Healthy Smart City via the Layered Architecture and DTRSC Tool

To support the transition of urban environments into adaptive and inclusive healthy SCs, we introduce the DTRSC tool, an open-source, modular platform that follows the proposed layered architecture.

Although some recent studies have proposed conceptual frameworks and architectural models for DTs in SCs and healthcare contexts, many do not offer practical tool realisations. For instance, Brucherseifer et al. [69] present a DT framework focused on infrastructure resilience, whereas Laamarti et al. [14] develop a standards-based DT model for health monitoring in urban settings. Similarly, Dihan et al. [70] provide a comprehensive review of DT architectures, implementation dimensions and future trends. These studies make valuable contributions, yet they do not provide modular deployable open-source software tools that cities can use directly with reasonable adaptation. Unlike other works, this research presents a simplified layered architecture and its practical realisation through the DTRSC tool. The platform is designed to be extensible, enabling municipal authorities or domain experts to plug in additional modules based on their specific application needs, making it suitable for a wide range of SC contexts. Its open-source nature further enhances flexibility and robustness, allowing for continuous refinement and customisation. Through the presented case studies, we demonstrate the tool's multi-domain capability, particularly its ability to integrate heterogeneous environmental and eHealth data for generating actionable insights for urban health development.

Kuru [33] highlights that SCs should provide tools for citizens to interact with events in their environment and participate in decision-making. This idea supports more inclusive and responsive city planning. Our DTRSC tool advances toward realising this vision by offering a practical tool designed to enhance the everyday lives of citizens through real-time insights, the integration of health and environmental data and modular service-related components that can be adapted to community needs. One of the core features of our platform is the integration of eHealth data, enabling population health modelling and scenario-based health assessments. This supports the discussion of Papachristou et al. [71] that DTs hold significant promise for advancing personalised urban healthcare planning.

The DTRSC tool can be effectively deployed by municipal authorities by integrating diverse data sources within an SC ecosystem. These sources may include a wide variety of sensor devices or IoT nodes, such as fixed environmental sensors (e.g., air or water quality monitors), mobile sensor nodes and wearable health devices used by city residents. Sensor devices and/or IoT nodes can be configured to run MQTT protocol clients (or other communication protocols, such as CoAP) and transmit their data over a secure Wi-Fi network provided by the SC municipality.

On the other hand, eHealth data can be collected directly from citizens' wearable devices via a data collection application (running on their smartphones), provided by the municipality for city residents, who create authorised accounts and consent to send their health data. These accounts can have specific system IDs, for which the mapping between the city resident's legal contact information and these IDs is kept confidential. eHealth data can also be securely shared by healthcare institutions, via a similar data collection application, for residents who have given their consent and can be transmitted directly and securely to city municipalities for integration into the system, with full respect for user anonymity and privacy, as these data should be labelled with the same system IDs. The data collection applications can also contain an MQTT client or other communication protocols as they interface with the implemented data transfer module in the DTRSC tool. The implementation of data collection applications and their security features is left to SC officials, as it must adhere to data sharing policies that may vary across different SCs and their affiliated health institutions. With informed consent, such data can also be stored to support training of machine learning models for predictive analytics and decision support.

The DTRSC tool can be deployed in a data centre managed by each SC municipality. This data centre contains edge servers that run the different modules of the tool as described in Section 4. The big data platform can also be deployed on these servers, allowing for the creation and training of ML models on the same infrastructure. After the models are trained, the data can be migrated to cloud servers to give space for the new incoming data that can be used to refine the trained models. This facilitates real-time data ingestion via big data technologies, enabling a continuous and dynamic flow of information into the DT system. Through its data processing and analytics module, the tool supports timely and context-aware decision-making by city officials. Additionally, the data stored in the cloud can be shared with other cities or partially retrieved as needed for training ML models or emulation purposes.

A key strength of our platform lies in its ability to correlate data from different urban domains. For instance, consider a scenario where data on water quality measurements and health condition indicators are both available. The tool, through appropriately designed ML models, can assess whether variations in water contaminants are potentially affecting community health outcomes. By enabling such cross-domain integration and real-time responsiveness, the DTRSC tool can serve as a central decision-support interface for municipalities, delivering tailored alerts and recommendations through a custom-designed portal, thereby contributing to an improvement in the QoL of city residents. The system's modular and extensible design ensures that it can evolve alongside city needs, making it a sustainable and impactful component of future SC initiatives.

Consequently, our implementation can link environmental, demographic and health data to enable a holistic view of public health in urban settings. Together, these features establish a practical and extensible foundation for transforming a city into a sustainable SC, grounded in real-world usability, health integration and open governance principles. Indeed, this can be achieved thanks to the wide availability of affordable smart wearables among city residents, which can effectively measure and record vital health signs, and the willingness of the city residents to share their measured data for the efficient operation of the digital health twin.

Limitations and Possible Mitigations

Although the DTRSC tool provides a foundational framework, several key considerations guide its implementation. As urban systems become increasingly interconnected and may exchange sensitive citizen data, security remains an essential concern. Rather than embedding hard-coded security policies and protocols, which often vary across cities, the tool offers open-ended flexibility for municipalities or authorities to incorporate their respective security standards and protocols, such as those outlined in city-specific privacy frameworks, emphasised by Kaya [33]. Another limitation is the current exclusion of automated cybersecurity features. As discussed by Kaya [33], immersive platforms, such as urban metaverses, require advanced security paradigms, including real-time anomaly detection and identity protection. These features have been intentionally left open for integration in future iterations, recognising that city administrators should tailor them based on local legislation and policy priorities.

Another consideration is scaling the tool in densely populated cities, where syncing data, handling high volumes and ensuring smooth integration across systems can be challenging. The current version of the tool has two limitations when handling a very large number of sensor devices or IoT nodes. The first one is the support of multiple MQTT brokers, whereas the second is the lack of remote management for MQTT clients. Both limitations can be mitigated in future updates by considering the automatic creation of a new broker configuration when the number of communicating devices exceeds a certain threshold and by developing an extension that allows remote access and management of the MQTT clients for administrative purposes. Additionally, further updates may add other features such as plugin-based analytics and support for collaboration between cities. Moreover, a more portable client interface can be developed to enhance accessibility and engagement for various stakeholders.

The tool's open architecture, compatibility with DT standards, and citizen-centric extensibility offer a promising direction, empowering stakeholders to customise and scale the platform according to the evolving demands of sustainable inclusive SCs.

Conclusion and Future Directions

This study tackles the practical introduction of a city DT to help transform cities into SCs with a focus on urban health development. Unlike other existing studies that focus primarily on architectural frameworks or specific use cases, our work provides both a simplified layered architecture and its practical realisation as a functional tool to support city planners and researchers in harnessing the DTs for SCs. Moreover, the proposed architecture and tool design facilitate the creation of DTs that integrate heterogeneous domains, such as real-time environmental data and residents' eHealth information, to provide intelligent analytics for generating actionable insights in a straightforward and practical manner. Additionally, the open-source nature of the tool ensures that others can easily adapt, extend or test new SC applications following the proposed layered architecture.

This architecture breaks down SC DT functionality into six layers, allowing for scalable real-time data emulation, data transfer, data ingestion, intelligent analytics, decision-making support and portable interaction for users or authorities. The DT architecture is envisioned through the proposed open-source DTRSC tool, which supports scalable real-time data generation, handling and analysis.

The tool can handle real-time data from physical devices and features a built-in data emulation module that generates correlated realistic data from virtual devices via a single or batch device network configuration. The tool echoes the proposed architecture by streaming the DT data to a big data platform that keeps track of the historical data, enabling longitudinal analysis. Moreover, the tool allows ML training and testing via multiple available techniques with easy extension to others, whereas decision support outcomes can be obtained through a simple user interface.

The practical capabilities of the DTRSC tool have been demonstrated through three case studies highlighting the role of DTs in enhancing public welfare. The first case, air quality monitoring, uses regression models to assess AQI levels, providing real-time insights into a SC's environmental conditions. In the second case, respiratory eHealth data from wearable devices are assumed to be transmitted to the DT, received and processed using classification models to assess individuals' respiratory conditions. The third case integrates air quality and wearable eHealth metrics data to implement a warning system that can offer a smart alert to city residents based on their health metrics and current AQI levels, showcasing the DTRSC tool's potential in real-world SC applications that combine environmental and eHealth data. These case studies also illustrate that authorities or government officials can operate the DTRSC tool, allowing users with limited programming knowledge to utilise its functionality.

Despite these advancements, certain limitations remain to be addressed in future work. The current tool does not include automated cyber security mechanisms, as these are expected to vary with city policies and are left open for future customisation. Future enhancements can also explore remote management of MQTT clients, plugin-based analytics, support for collaboration between cities and extending interactions beyond human users to include cyber–physical systems.

Author Contributions

N. Jayachandran contributed to developing the DT architecture and the proposed tool. She also contributed to paper writing.A. Abdrabou contributed to the design of the proposed architecture and tool. He also participated in writing the paper. M. AlBataineh and K. Noordin contributed to reviewing and writing the paper.

Acknowledgements

This work was made possible by the UAE University AUA Grant 12N143.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available in [Microsoft Research] at , in [Reddy M. Wearables Dataset 2023] at and in [Health Monitoring System Dataset] at .

References

T. Deng, K. Zhang, and Z. J. M. Shen, “A Systematic Review of a Digital Twin City: A New Pattern of Urban Governance Toward Smart Cities,” Journal of Management Science and Engineering 6, no. 6 (2021): 125–134, https://doi.org/10.1016/j.jmse.2021.03.003.

S. Ivanov, K. Nikolskaya, G. Radchenko, L. Sokolinsky, and M. Zymbler, “Digital Twin of City: Concept Overview,” in 2020 Global Smart Industry Conference (GloSIC) (IEEE, 2020), 178–186, https://doi.org/10.1109/GloSIC50886.2020.9267879.

A. V. Mukhacheva, M. N. Ugryumova, I. S. Morozova, and M. Y. Mukhachyev, “Digital Twins of the Urban Ecosystem to Ensure the Quality of Life of the Population,” in Proceedings of the International Scientific and Practical Conference Strategy of Development of Regional Ecosystems “Education‐Science‐Industry” (ISPCR 2021) (Atlantis Press, 2022): 331–338, https://doi.org/10.2991/aebmr.k.220208.047.

D. M. Botín‐Sanabria, A. S. Mihaita, R. E. Peimbert‐García, M. A. Ramírez‐Moreno, R. A. Ramírez‐Mendoza, and JdJ. Lozoya‐Santos, “Digital Twin Technology Challenges and Applications: A Comprehensive Review,” Remote Sensing 14, no. 6 (2022), https://www.mdpi.com/2072‐4292/14/6/1335.

G. del Campo, E. Saavedra, L. Piovano, F. Luque, and A. Santamaria, “Virtual Reality and Internet of Things Based Digital Twin for Smart City Cross‐Domain Interoperability,” Applied Sciences 14, no. 7 (2024): 2747, https://www.mdpi.com/2076‐3417/14/7/2747.

S. Shen, “Construction of Smart Transportation City System Based on Digital Twins. Advances in Transdisciplinary Engineering,” in Intelligent Transportation and Smart Cities,” (2024), Open: Access under CC BY‐NC 4.0., https://doi.org/10.3233/ATDE240377.

F. Dembski, U. Wössner, M. Letzgus, M. Ruddat, and C. Yamu, “Urban Digital Twins for Smart Cities and Citizens: The Case Study of Herrenberg, Germany,” Sustainability 12, no. 6 (2020): 2307, https://www.mdpi.com/2071‐1050/12/6/2307.

Y. Shamlitsky, O. Aleksey, E. Morozov, T. Strekaleva, “Using Digital Twins to Manage Traffic Flows.” in E3S Web of Conferences 471 (2024): 04027. Presented at TITDS‐XIV‐2023: 14th International Scientific and Practical Conference on Transport Infrastructure: Territory Development and Sustainability, https://doi.org/10.1051/e3sconf/202447104027.

T. Rantanen, A. Julin, J. P. Virtanen, H. Hyyppä, and M. T. Vaaja, “Open Geospatial Data Integration in Game Engine for Urban Digital Twin Applications,” ISPRS International Journal of Geo‐Information 12, no. 8 (2023): 310, https://www.mdpi.com/2220‐9964/12/8/310.

M. N. Kamel Boulos and P. Zhang, “Digital Twins: From Personalised Medicine to Precision Public Health,” Journal of Personalized Medicine 11, no. 8 (July 2021): 745, https://doi.org/10.3390/jpm11080745.

R. D’Hauwers, N. Walravens, and P. Ballon, “From an Inside‐In towards an Outside‐Out Urban Digital Twin: Business Models and Implementation Challenges,” ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences VIII‐4/W1–2021 (2021): 25–32, https://doi.org/10.5194/isprs‐annals‐viii‐4‐w1‐2021‐25‐2021

S. Vempati, “Securing Smart Cities: A Cybersecurity Perspective on Integrating IoT, AI, and Machine Learning for Digital Twin Creation,” Journal of Electrical Systems 20, no. 5 (2024): 2817–2827, https://doi.org/10.52783/jes.3052.

M. Hämäläinen, “Urban Development With Dynamic Digital Twins in Helsinki City,” IET Smart Cities 3, no. 4 (December 2021): 201–210, https://doi.org/10.1049/smc2.12015.

F. Laamarti, H. F. Badawi, Y. Ding, F. Arafsha, B. Hafidh, and A. E. Saddik, “An ISO/IEEE 11073 Standardized Digital Twin Framework for Health and Well‐Being in Smart Cities,” IEEE Access 8 (2020): 105950–105961, https://doi.org/10.1109/access.2020.2999871.

W. O. N. –. d’Hollosy, L. van Velsen, A. Henket, and H. Hermens, “An Interoperable eHealth Reference Architecture for Primary Care,” in 2018 IEEE Symposium on Computers and Communications (ISCC) (IEEE, 2018): 01090–01095, https://doi.org/10.1109/ISCC.2018.8538576.

Z. Allam and D. S. Jones, “Future (Post‐COVID) Digital, Smart and Sustainable Cities in the Wake of 6G: Digital Twins, Immersive Realities and New Urban Economies,” Land Use Policy 2 (2021): 101.

G. Ahmadi‐Assalemi, H. Al‐Khateeb, and A. Aggoun, “Privacy‐Enhancing Technologies in the Design of Digital Twins for Smart Cities,” Network Security 2022, no. 7 (July 2022): S1353–S4858: (22)70046–3, https://doi.org/10.12968/s1353‐4858(22)70046‐3.

H. Rafik, A. Maizate, and A. Ettaoufik, “Data Security Mechanisms, Approaches, and Challenges for E‐Health Smart Systems,” International Journal of Online and Biomedical Engineering (iJOE) 19, no. 2 (February 2023): 42–66, https://online‐journals.org/index.php/i‐joe/article/view/37069.

I. Jeddoub, G. A. Nys, R. Hajji, and R. Billen, “Digital Twins for Cities: Analyzing the Gap Between Concepts and Current Implementations With a Specific Focus on Data Integration,” International Journal of Applied Earth Observation and Geoinformation 8 (2023): 122.

H. Wang, X. Chen, F. Jia, and X. Cheng, “Digital Twin‐Supported Smart City: Status, Challenges and Future Research Directions,” Expert Systems With Applications 5 (2023): 217.

M. Attaran and B. G. Celik, “Digital Twin: Benefits, Use Cases, Challenges, and Opportunities,” Decision Analytics Journal 3 (2023): 6, https://doi.org/10.1016/j.dajour.2023.100165.

M. Ziehl, R. Herzog, T. Degkwitz, M. H. Niggemann, G. Ziemer, and R. Thoneick, “Transformative Research in Digital Twins for Integrated Urban Development: Two Real‐World Experiments on Unpaid Care Workers Mobility,” International Journal of E‐Planning Research 12, no. 1 (2023): 1–18, https://doi.org/10.4018/ijepr.333851.

Y. Huang, H. Peng, M. Sofi, et al., “The City Management Based on Smart Information System Using Digital Technologies in China,” IET Smart Cities 4, no. 3 (September 2022): 117–126, https://doi.org/10.1049/smc2.12035.

G. Schrotter and C. Hürzeler, “The Digital Twin of the City of Zurich for Urban Planning. PFG—Journal of Photogrammetry,” Remote Sensing and Geoinformation Science 88 (2020): 99–112.

Y. B. Ravid and M. Aharon‐Gutman, “The Social Digital Twin: The Social Turn in the Field of Smart Cities,” Environment and Planning B: Urban Analytics and City Science 50, no. 6 (2022): 1455–1470, https://doi.org/10.1177/23998083221137079.

T. Nochta, L. Wan, J. M. Schooling, and A. Akp, “Socio‐Technical Perspective on Urban Analytics: The Case of City‐Scale Digital Twins,” Journal of Urban Technology 28, no. 1–2 (2021): 263–287, https://doi.org/10.1080/10630732.2020.1798177.

S. Shaharuddin, K. N. Abdul Maulud, S. A. F. Syed Abdul Rahman, A. I. Che Ani, “DIGITAL TWIN FOR INDOOR DISASTER IN SMART CITY: A SYSTEMATIC REVIEW,” International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 2022 XLVI‐4/W3‐2021: 315–322, https://isprs‐archives.copernicus.org/articles/XLVI‐4‐W3‐2021/315/2022/.

D. N. Ford and C. M. Wolf, “Smart Cities With Digital Twin Systems for Disaster Management,” Journal of Management in Engineering 36, no. 4 (2020): ORCID: David N. Ford ‐ 0000‐0003‐3511‐1360, https://doi.org/10.1061/(ASCE)ME.1943‐5479.0000779.

M. R. M. F. Ariyachandra and G. Wedawatta, “Digital Twin Smart Cities for Disaster Risk Management: A Review of Evolving Concepts,” Sustainability 15, no. 15 (2023): 11910, https://www.mdpi.com/2071‐1050/15/15/11910.

V. Astarita, G. Guido, S. S. Haghshenas, and S. S. Haghshenas, “Risk Reduction in Transportation Systems: The Role of Digital Twins According to a Bibliometric‐Based Literature Review,” Sustainability 16, no. 8 (2024): 3212, https://www.mdpi.com/2071‐1050/16/8/3212.

H. Yeon, T. Eom, K. Jang, and J. Yeo, “DTUMOS, Digital Twin for Large‐Scale Urban Mobility Operating System,” Scientific Reports 13, no. 1 (2023): 5154, https://doi.org/10.1038/s41598‐023‐32326‐9.

D. Szpilko, X. Fernando, E. Nica, K. Budna, A. Rzepka, and G. Lăzăroiu, “Energy in Smart Cities: Technological Trends and Prospects,” Energies 17, no. 24 (2024): 6439, https://doi.org/10.3390/en17246439.

K. Kuru, “MetaOmniCity: Toward Immersive Urban Metaverse Cyberspaces Using Smart City Digital Twins,” IEEE Access 11 (2023): 43844–43868, https://doi.org/10.1109/access.2023.3272890.

J. Van der Ven, B. J. F. Van den Bemt, L. Van Dijk, et al., “Preferences of Patients With Musculoskeletal Disorders Regarding the Timing and Channel of eHealth and Factors Influencing Its Use: Mixed Methods Study,” JMIR Human Factors 10 (2023): e44885, https://doi.org/10.2196/44885.

Ph Huang, Kh Kim, and M. Schermer, “Ethical Issues of Digital Twins for Personalized Health Care Service: Preliminary Mapping Study,” Journal of Medical Internet Research 24, no. 1 (January 2022): e33081, https://doi.org/10.2196/33081.

E. Ferko, A. Bucaioni, and M. Behnam, “Architecting Digital Twins,” IEEE Access 10 (2022): 50335–50350, https://doi.org/10.1109/access.2022.3172964.

G. Lăzăroiu, T. Gedeon, E. Rogalska, et al., “The Economics of Deep and Machine Learning‐Based Algorithms for COVID‐19 Prediction, Detection, and Diagnosis Shaping the Organizational Management of Hospitals,” Oeconomia Copernicana 15, no. 1 (2024): 27–58, https://doi.org/10.24136/oc.2984.

C. Qian, X. Liu, C. Ripley, M. Qian, F. Liang, and W. Yu, “Digital Twin—Cyber Replica of Physical Things: Architecture, Applications and Future Research Directions,” Future Internet 14, no. 2 (2022): 64, https://www.mdpi.com/1999‐5903/14/2/64.

D. Szpilko, F. J. Naharro, G. Lazaroiu, E. Nica, and A. D. L. T. Gallegos, “Artificial Intelligence in the Smart City — A Literature Review,” Engineering Management in Production and Services 15, no. 4 (December 2023): 53–75, https://doi.org/10.2478/emj‐2023‐0028.

K. Kantawong, S. Chaichumpa, S. Pravesjit, and M. Yaibuates, “A Lightweight Framework for Retrieve IP Device Status Based on MQTT Protocol,” in 2022 Joint International Conference on Digital Arts, Media and Technology with ECTI Northern Section Conference on Electrical, Electronics, Computer and Telecommunications Engineering (ECTI DAMT & NCON) (IEEE, 2022): 46–49, https://doi.org/10.1109/ECTIDAMTNCON53731.2022.9720332.

C. Bormann, A. P. Castellani, and Z. Shelby, “CoAP: An Application Protocol for Billions of Tiny Internet Nodes,” IEEE Internet Computing 16, no. 2 (2012): 62–67, https://doi.org/10.1109/mic.2012.29.

K. Newt. PyShark. PyPI 2018, https://pypi.org/project/pyshark/.

Foundation AS. Apache Kafka Apache (2020), https://kafka.apache.org/.

World Health Organization, Ambient Air Pollution: A Global Assessment of Exposure and Burden of Disease (World Health Organization, 2024), https://www.who.int/health‐topics/air‐pollution#tab=tab_1.

E. Marino, M. Caruso, D. Campagna, and R. Polosa, “Impact of Air Quality on Lung Health: Myth or Reality?,” Therapeutic Advances in Chronic Disease 6, no. 5 (2015): 286–298, https://doi.org/10.1177/2040622315587256.

Y. Zheng, X. Yi, M. Li, et al., “Forecasting Fine‐Grained Air Quality Based on Big Data,” in Proceedings of the 21st ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (ACM, 2015), 2267–2276, https://doi.org/10.1145/2783258.2788573.

U S Environmental Protection Agency, Technical Assistance Document for the Reporting of Daily Air Quality – the Air Quality Index (AQI) (Office of Air Quality Planning and Standards, 2024): Air Quality Assessment Division. EPA‐454/B‐24‐002, https://www.airnow.gov/publications/air‐quality‐index/technical‐assistance‐document‐for‐reporting‐the‐daily‐aqi/.

A. Yadav, C. K. Jha, and A. Sharan, “Optimizing LSTM for Time Series Prediction in Indian Stock Market,” Procedia Computer Science 167 (2020): 2091–2100, https://doi.org/10.1016/j.procs.2020.03.257.

A. K. Dubey, A. Kumar, V. García‐Díaz, A. K. Sharma, and K. Kanhaiya, “Study and Analysis of SARIMA and LSTM in Forecasting Time Series Data,” Sustainable Energy Technologies and Assessments 47 (2021): 101474, https://doi.org/10.1016/j.seta.2021.101474.

A. V. Tatachar, “Comparative Assessment of Regression Models Based on Model Evaluation Metrics,” International Journal of Innovative Technology and Exploring Engineering 8, no. 9 (2021): 853–860.

D. Chicco, M. J. Warrens, and G. Jurman, “The Coefficient of Determination R‐Squared Is More Informative Than SMAPE, MAE, MAPE, MSE and RMSE in Regression Analysis Evaluation,” PeerJ Computer Science 7 (2021): e623, https://doi.org/10.7717/peerj‐cs.623.

A. Zanella, N. Bui, A. Castellani, L. Vangelista, and M. Zorzi, “Internet of Things for Smart Cities,” IEEE Internet of Things Journal 1, no. 1 (2014): 22–32, https://doi.org/10.1109/jiot.2014.2306328.

B. A. Vicente, R. Sebastião, and V. Sencadas, “Wearable Devices for Respiratory Monitoring,” Advanced Functional Materials 34, no. 45 (November 2024): 2404348, https://advanced.onlinelibrary.wiley.com/doi/10.1002/adfm.202404348.

I. C. Jeong, D. Bychkov, and P. C. Searson, “Wearable Devices for Precision Medicine and Health State Monitoring,” IEEE Transactions on Biomedical Engineering 66, no. 5 (May 2019): 1242–1258, https://ieeexplore.ieee.org/document/8470151/.

D. Liaqat, M. Abdalla, P. Abed‐Esfahani, et al., “WearBreathing: Real‐World Respiratory Rate Monitoring Using Smartwatches,” Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 3, no. 2 (2019): 1–22, https://doi.org/10.1145/3328927.

M. Reddy Mandhadi, Wearables Dataset (Kaggle, 2023), https://www.kaggle.com/datasets/manideepreddy966/wearables‐dataset?select=personal_health_data.csv.

Garmin, “The Critical Relationship Between Respiratory Rate, Heart Rate, and Cadence” (Garmin, 2024), https://www.garmin.com/en‐PH/blog/critical‐relationship‐between‐respiratory‐rate‐heart‐rate‐and‐cadence/.

H.‐K. Ra, A. Salekin, H. J. Yoon, J. Kim, et al., “Asthmaguide: An Asthma Monitoring and Advice Ecosystem,” in 2016 IEEE Wireless Health (WH) (IEEE, 2016): 1–8, https://doi.org/10.1109/WH.2016.7764567.

Vakili M., Ghamsari M., Rezaei M., “Performance Analysis of Machine Learning Models for IoT Data Classification.” arXiv preprint arXiv:220209636 (2022).

N. R. Bommela, Health Monitoring System Dataset (Kaggle, 2024), https://www.kaggle.com/datasets/nraobommela/health‐monitoring‐system?select=what‐is‐a‐normal‐respiratory‐rate‐2248932‐v1‐5c1abe6846e0fb0001c6284a.png.

T. Götschi, J. Heinrich, J. Sunyer, and N. Künzli, “Long‐Term Effects of Ambient Air Pollution on Lung Function: A Review,” Epidemiology 19, no. 5 (2008): 690–701, https://doi.org/10.1097/ede.0b013e318181650f.

K. Kuru and D. Ansell, “TCitySmartF: A Comprehensive Systematic Framework for Transforming Cities into Smart Cities,” IEEE Access 8 (2020): 18615–18634, https://doi.org/10.1109/access.2020.2967777.

S. Mazzetto, “A Review of Urban Digital Twins Integration, Challenges, and Future Directions in Smart City Development,” Sustainability 16, no. 19 (2024): 8337, https://doi.org/10.3390/su16198337.

J. Argota Sánchez‐Vaquerizo. Urban Digital Twins and Metaverses Towards City Multiplicities: Uniting or Dividing Urban Experiences? Ethics and Information Technology. 27, no. 4 (2025).

Z. Lv, W. L. Shang, and M. Guizani, “Impact of Digital Twins and Metaverse on Cities: History, Current Situation, and Application Perspectives,” Applied Sciences 12, no. 24 (December 2022): 12820, https://www.mdpi.com/2076‐3417/12/24/12820.

G. White, A. Zink, L. Codecá, and S. Clarke, “A Digital Twin Smart City for Citizen Feedback,” Cities 110 (March 2021): 103064, https://linkinghub.elsevier.com/retrieve/pii/S0264275120314128.

M. Aloqaily, O. Bouachir, F. Karray, I. Al Ridhawi, and A. E. Saddik, “Integrating Digital Twin and Advanced Intelligent Technologies to Realize the Metaverse,” IEEE Consumer Electronics Magazine 12, no. 6 (November 2023): 47–55, https://ieeexplore.ieee.org/document/9913665/.

O. Krylov, “Designing a New Digital Twin Framework for,” Smart Cities (2025), https://ssrn.com/abstract=5100117.

E. Brucherseifer, H. Winter, A. Mentges, M. Mühlhäuser, and M. Hellmann, “Digital Twin Conceptual Framework for Improving Critical Infrastructure Resilience,” At ‐ Automatisierungstechnik 69, no. 12 (December 2021): 1062–1080, https://doi.org/10.1515/auto‐2021‐0104.

M. S. Dihan, A. I. Akash, Z. Tasneem, et al., “Digital Twin: Data Exploration, Architecture, Implementation and Future,” Heliyon 10, no. 5 (2024): e26503, https://doi.org/10.1016/j.heliyon.2024.e26503.

K. Papachristou, P. F. Katsakiori, P. Papadimitroulas, L. Strigari, and G. C. Kagadis, “Digital Twins’ Advancements and Applications in Healthcare, Towards Precision Medicine,” Journal of Personalized Medicine 14, no. 11 (2024): 1101, https://doi.org/10.3390/jpm14111101.