IoT-Enabled Machine Learning for Comprehensive Water Quality Assessment in the Mahanadi River: A Multibelt Analysis of Seasonal Contamination and Predictive Modeling

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

1. Introduction

Rivers serve as essential ecological support systems, sources of power, and convenient means for human transport and communication. They provide the raw materials and resources necessary to sustain human life. Furthermore, rivers are vital sources of life for numerous towns and villages across the country. With the growth of industrial zones and the expansion of cities, the Mahanadi River struggles to maintain clean water. Natural phenomena and human activities have exacerbated this situation, making it challenging for environmental workers to meet the increasing demands of a growing population [1]. This has led to numerous facets of water contamination. The intricate interactions among natural processes, industrial discharges, household effluents, and agricultural runoff contribute to this mosaic of problems [2]. Monitoring river water has become increasingly imperative to ensure environmental safety. The intricate interplay between pollutants evaluates the quality of the water. Traditional methods require a systematic and up-to-date grasp at all times. Accordingly, the dividing line between home and away can, in fact, connect when cutting-edge technology is used to bridge that divide. It is of utmost importance that we use up-to-date technologies to manage this indispensable resource on a sustainable basis. The river, often regarded as the life of the region, plays a crucial role in the lives and ecosystems of people living downstream. The Mahanadi River is rich in biological resources and supports vast fertile plains, which play a crucial role in the region’s agricultural productivity. These two factors are critical to the socioeconomic development of this region. The Mahanadi River is significant, but it faces increasingly severe problems, such as water pollution. Several factors contribute to the decline in water quality. These include industrial effluent, untreated wastewater from households, and agricultural runoff (which comes with pesticides and fertilizers), plus natural causes [3]. Deterioration threatened the ecological balance of the river and even threatened the peripheral people who depended on it. However, all these aspects of rapid urbanization and industrialization leave no room for delay. There is an urgent need for high-tech monitoring methods to ensure real-time data on critical water quality parameters. These parameters are essential indicators of overall water health. pH influences biological processes and determines the solubility of nutrients and heavy metals, affecting aquatic ecosystems [2]. Dissolved oxygen (DO) is crucial for sustaining aquatic life, while biochemical oxygen demand (BOD) and chemical oxygen demand (COD) quantify organic and inorganic pollutants, respectively [4]. Total coliforms (TCs) serve as microbial contamination indicators and are essential for assessing public health risks [3]. Monitoring these parameters enables early detection of pollution and aids in sustainable water management [5]. Good environmental, social, and economic health is inseparable from monitoring the Mahanadi River. The Mahanadi River is closely related to environmental, social, and economic health [6]. Clear water is maintained so there is enough supply for human beings without becoming toxic while containing diseases and preserving the richness of life. Furthermore, monitoring provides information on the impact of climate change, can serve as a basis for adaptation measures, and can reduce the effects on ecosystems [7]. This monitoring protects human health by certifying that potable water meets every applicable safety standard [8]. Monitoring of the Mahanadi River is crucial to sustainable development, habitat preservation, and the well-being of the region.

This research is aimed at helping resolve the limitations inherent in conventional water quality assessment methods using Internet of Things (IoT)–enabled monitoring systems and next-generation machine learning models. The study uses advanced data collection technology and machine learning techniques to analyze how water pollution on the Mahanadi River differs by location, seasonality, and belts.

I. The main goal is to implement and integrate cutting-edge IoT technology to monitor pH, DO, BOD, COD, and TC in real time [9, 10].

II. The research is aimed at creating and deploying advanced predictive models on real-time data.

III. The comparative analysis of water quality throughout various seasons and geographical regions along the Mahanadi River constitutes.

IV. Calculating a comprehensive water quality index (WQI) for all seasons and regions.

V. Examine intricate connections between water quality measures, geographical regions, and seasonal fluctuations utilizing cutting-edge analytical methods.

We provide some fundamental ideas to understand and control the water quality of the Mahanadi River with our holistic methodology. With the help of IoT technology, data can be captured in real time. The frequency and granularity achieved by monitoring critical water parameters are higher than average. If we apply predictive modeling to this live dataset, our ability to forecast changes in water quality will be significantly improved. Furthermore, having this capability allows us to exercise some degree of control based on prediction and a high level of anticipation accuracy. Checking the WQI in various geographical regions and seasons also tells us much about their intricate interplay. WQI is a valuable method to ensure that water purity indicators can be united and standardized when comparing differences in water quality between different locations, seasons, or belts. This research combines these contributions to describe the dynamics of the water quality of the Mahanadi River.

Consequently, such work forms the basis for policymaking, environmental protection, and ecologically sustainable management. This research contributes to environmental science by providing helpful information to those working to conserve and restore the Mahanadi River ecosystem. Our research helps preserve and recover the Mahanadi River environment for humankind forever.

2. Related Work

Excessive consumption of land and sea resources, coupled with the intensifying effects of globalization, has contributed to a decrease in the purity of potable water. Additionally, the increase in population has placed significant pressure on the water systems that supply communities, further exacerbating the scarcity caused by the drying up of rivers due to the effects of global climate change. Immediate implementation of intelligent river water management systems is necessary to ensure robust surveillance of the quality and quantity of drinking water to address these urgent challenges. A water monitoring system has become essential to our daily lives, as it enables real-time control over water quality and facilitates prudent allocation of management resources in urban areas.

We performed an exhaustive analysis of the numerous systems that researchers developed to produce a model of superior quality. Although focusing on parameters including temperature, pH, conductivity, and others, many authors have proposed novel models for evaluating water quality. We devised an intelligent river water monitoring system to carry out each monitoring responsibility in light of these observations.

Remote sensing imaging has been used alongside machine learning algorithms, IoT, and wireless sensors in the current literature to forecast pollution levels in lagoon waters. This novel methodology allows continuous monitoring of contaminants, with possible implementations in agriculture [11]. Previous studies employed machine learning techniques to forecast water quality but needed help obtaining adequate results. A study suggested a technique for monitoring contamination in real time by using sensors to measure characteristics such as temperature and water discharge. However, this approach is only applicable to a particular geographic region. Dissolved solid analysis (DSA) was performed to examine water contaminants in single and mixed water samples by Smith et al. [12]. Several studies have demonstrated the effectiveness of AI and IoT-based water quality monitoring methods. Li et al. [11] achieved an accuracy of 92% in predicting water quality using machine learning models, whereas Cui et al. [13] developed an IoT-based microbial contamination detection system with a sensitivity rate of 95% [14–16]. Porwal et al. [17] utilized support vector machines (SVMs) for water quality classification, achieving an $F1$ score of 0.88, indicating high classification reliability. While these studies highlight the advantages of AI and IoT, challenges remain in terms of adaptability across diverse water bodies, sensor calibration, and data scalability [17]. Real-time awareness is considered essential in flood-prone areas such as the Philippines, leading to the development of advanced flood monitoring and forecasting systems by Lee et al. [18]. An economically efficient supervisory control and data acquisition (SCADA) system utilizing a GSM module was suggested to remotely monitor water quality indicators by Johnson et al. [19]. Novel instruments, including Forel-Ule color scale stickers, the TurbAqua mobile app, and integrated platinum (Pt) sensors in a microelectromechanical system (MEMS), were developed to detect water color and clarity [20, 21]. A real-time turbidity monitoring system utilizing infrared detectors was developed as an open-source optical system by Rao et al. [22]. In addition, a microcontroller system was created to communicate wirelessly and measure water quality parameters. This system demonstrates a high level of sensitivity and accuracy, which is comparable to commercial sensors by Cui et al. and Alam et al. [13, 23]. Khan et al. [24] investigated the seasonal fluctuations and clustering of monitoring stations along the Yamuna and Ganga rivers in the Indian state of Uttarakhand using cluster analysis (CA) components, principal component analysis (PCA), and correlation analysis.

Figure 1 highlights that the full potential of these technologies, especially ML, in river water monitoring still needs to be realized, even though research trends point to steady growth in exploring RWMS with IoT. The research community is currently focused on using connectivity and real-time data capture, as seen by the significant influence of IoT in recent years. The dynamic character of these trends highlights the necessity of ongoing research and development in the RWMS field using IoT and machine learning.

[figure(s) omitted; refer to PDF]

3. Identification of Water Quality in the Odisha Section of the Mahanadi River

The monitoring stations were strategically placed to capture diverse pollution sources and environmental conditions along the Mahanadi River. Instead of uniform placement, the stations were chosen based on key environmental and geographical factors, ensuring representation of distinct belt types (industrial, residential, agricultural) and varying pollution intensities. This approach allows for a more comprehensive assessment of seasonal contamination patterns, enabling precise water quality monitoring in areas most impacted by human and industrial activities [25–27]. Traditional water quality testing techniques are isolated, complicating effective water environment management.

This study introduces a water quality monitoring system for the Mahanadi River, leveraging an intelligent decision-making module based on IoT architecture. This system enhances river environment management by enabling automatic and intelligent data collection for water quality analysis [28]. The main objective is to implement a capable mechanism to continuously monitor river water quality. To improve the quality of water, this system focuses on assessing vital indicators, including pH, DO, BOD, COD, and TC.

3.1. IoT Architecture of the Monitoring System

This paper divides the IoT system into three parts: the physical, middleware, and application layer. The system implements the divide-and-conquer strategy in response to the diverse monitoring waters and complex environment [29, 30]. To do this, the entire water area is divided into 10 monitoring subnetworks. While operational, the monitoring terminal creates a water quality monitoring network in the designated water area. The water quality gateway then transmits the water quality data through the main transmission network. The monitoring center handles data aggregation and management, ultimately allowing online water quality monitoring of the designated water area. Figure 2 shows the complete design framework of the IoT water quality monitoring system.

[figure(s) omitted; refer to PDF]

Figure 2 captures the orchestrated data flow from the physical layer, through efficient transmission through middleware, to the cloud application for in-depth analysis. The user-friendly interface at the application layer ensures accessibility and usability, making the river water monitoring system a comprehensive and effective tool to manage the water quality of the Mahanadi River. The IoT monitoring architecture uses the water quality monitoring terminal. It is widely deployed in certain bodies of water, and its main function is to gather and transmit water quality information in real time [31]. The proposed monitoring system used Wi-Fi technology to establish subnetworks for dynamic monitoring of water quality. During operation, the terminals collect data on pH, DO, BOD, COD, and TC levels in the water. They follow the system instructions and aggregate significant water quality data within the network. Figure 3 illustrates a schematic diagram of the proposed RWMS.

[figure(s) omitted; refer to PDF]

Figure 3 visually represents the proposed IoT-based river water monitoring (IoT-RWM) system. The system was developed to collect data from the Mahanadi River in 10 distinct locations spanning different sectors (industrial, residential, and agricultural) and varying seasons (summer, winter, fall, and monsoon). The sensor deployment process culminates in the computation of the WQI via normalization and machine learning classification, as depicted in the schematic diagram. This IoT-RWM system is aimed at improving the monitoring and categorizing of water quality in the Mahanadi River, providing vital knowledge for efficient environmental governance.

3.2. Description of the Study Area

The Mahanadi River, a prominent river that flows in the eastern direction of Peninsular India, mostly passes through Chhattisgarh, Odisha, and, to a lesser degree, Jharkhand, Maharashtra, and Madhya Pradesh. The river originates in the Dhamtari district of Chhattisgarh at an elevation of around 442 m above sea level. It stretches for a distance of 851 km before reaching its endpoint in the Bay of Bengal. The Mahanadi River basin, which spans 141,600 km² or approximately 4% of the country’s total land area, is classified as a predominantly mixed coastal plain estuary due to its physical attributes. The basin has a tropical monsoon climate. The basin’s land is mostly used for agriculture, forest reserves, industries, and residential areas. Sampling sites were strategically placed across the basin to thoroughly evaluate the state of the Mahanadi River as it flows through the entire basin. An intensive effort was made to assess surface water quality in important places along the river as part of the sampling and analyzing of river water using IoT technologies. For the current investigation, water quality data obtained by the IoT node from 10 monitoring stations were utilized. These stations include Paradeep (S-1), Jharsugoda (S-2), Kathajodi (S-3), Sankhatras (S-4), Kuakhai (S-5), Kua R (S-6), Daya at kanas (S-7), Rajdhani College (S-8), Chandaka forest (S-9), and Khordha vadimula (S-10) shown in Figure 4 and the study area shown in Figure 5.

[figure(s) omitted; refer to PDF]

Every month for the entire year of 2022, water samples were taken below the surface. The sampling stations that were selected were arranged in a way that would cover a range of seasonal and belt changes. This all-encompassing sampling strategy sought to capture dynamic variations in water quality throughout the year, considering the distinctive qualities of each site and the impact of several seasons. Detailed studies and evaluations were based on data from the water parameters of the different stations. IoT technology guarantees precise and real-time monitoring of critical water quality indicators.

3.3. Experimental Setup

Real-time experiments were conducted for a year to assess the water quality of the Mahanadi River in Odisha. The evaluation included sensors to quantify different characteristics of water quality. The experimental setup utilized in the above experiments comprised several vital elements:

➢ A pH sensor is tasked with the critical responsibility of determining the chemical composition of a river by analyzing its pH level.

➢ The DO sensor monitors and quantifies the amount of oxygen in the water. Measurement of DO is a critical indicator of a river’s capacity to sustain aquatic life.

➢ The BOD sensor is determined by evaluating the amount of oxygen microorganisms consume as they decompose organic matter in water. This metric provides insight into the extent of waste contamination.

➢ The COD sensor determines the informational value of chemical contaminants present in water by quantifying the oxygen required for the chemical degradation of pollutants.

➢ The TC sensor detects the concentration of TC bacteria, an indicator of microbial contamination and the associated risks to human health.

The integration of the sensors produced a robust monitoring system that consistently assessed and tracked the dynamic changes in water quality over the specified duration of 1 year. A comprehensive understanding of the water quality of the Mahanadi River was achieved through real-time experiments, which is critical for environmental management and decision-making in the region. Table 1 shows the type of sensors and their measurement standards according to the Department of Water Resources & River Development. The sensors used in this study were calibrated to meet these standards, ensuring accuracy and consistency in data collection throughout the experimental setup.

Table 1

Sensors measuring standards of water (Indian).

Parameters	Sensor type	Range	Accuracy
pH	pH sensor	6.5–8.5	±0.01 pH
DO	DO sensor	> 4.0	±0.1 to ±0.2 mg/L
BOD	BOD sensor	< 3.0	±0.1 to ±0.5 mg/L
COD	COD sensor	Maximum 10	±5 to ±10
TC	Microbial sensor	< 5000	±2% or ±0.1 NTU

Data collection and transmission to the real-time database through the IoT cloud depended on the sensors.

3.4. Dataset Generation

The dataset was effectively collected using IoT sensors located at 10 individual monitoring points over 12 months. These sensors are aimed at collecting as much information about the parameters crucial for the quality of water, such as pH, DO, BOD, COD, and TC, a primary indicator in measuring biological pollutants, which can alter the chemical properties within an aquatic environment. Here, we are also interested. All stations in the data collection process incorporated continuous monitoring and recordkeeping of these parameters. This led to a coherent dataset that represents the fluctuation in water quality throughout the year in fine detail. Ten monitoring stations were established in residential, industrial, and agricultural areas along the banks of 10 angling grounds on the Mahanadi River. Data from the four seasons (summer, winter, autumn, and monsoon) were included in the dataset. The dataset covers different settings of changes in water quality measures. The long duration of the monitoring and various geographic locations led to a deep understanding of the different variations in water quality on the Mahanadi River. The dataset has 121 rows and 13 columns. Each row in the dataset means a different observation or measurement; each column refers to one of the variables studied. These accumulated data, including pH, DO, BOD, COD, and TC measurements, provide a solid foundation for in-depth analysis and assessment. This dataset is invaluable for understanding the environmental health of the Mahanadi River. To extract insights into trends, patterns, and potential relationships among water quality measures, it is only when we fully understand these elements that we can hold onto or add value; it has never been strictly about cost-cutting only.

3.5. WQI

The WQI is a numerical rating of water quality based on several standards that determine whether it should be used for drinking or washing. Different numerical ranges were used in previous research [32–34], for example, to represent the quality index. These ranges were based on the number of weights assigned to different criteria. The WQI was created by processing the dataset, assigning weights to each indicator and subsets of indicators, or calculating these inherent in individual indices.

3.5.1. Preprocessing

After data processing, we do min–max normalization. Normalization is the technique of making all attributes in the dataset using a standard scale. Naturally, the features of different ranges need to be stretched. Thus, it makes the model evaluation and prediction of water quality more sustainable and accurate. Equation (1) shows the calculated normalized values for each parameter: pH, DO, BOD, COD, and TC: $$\begin{array}{}\text{(1)}& N_i=\frac{\text{sensed}\text{value}{\text{Sv}}_i-\text{minimum}\text{value}{\min }_i}{\text{maximum}\text{value}{\max }_i-\text{minimum}\text{value}{\min }_i}.\end{array}$$

After establishing the normalized value for each parameter, we proceeded with assigning weights ($W_i$) to separate parameters in order to determine the subindex. The subindex ($S_i$) is determined by multiplying the weights allocated to each parameter (pH: 0.25, DO: 0.20, COD: 0.15, BOD: 0.15, TC: 0.25) with their respective normalized values ($N_i$) in Equation (2): $$\begin{array}{}\text{(2)}& {SI}_i=W_i-N_i.\end{array}$$

Once the subindex has been determined, the total subindex calculation is obtained by adding the various subindices for each parameter. The subindex for each parameter is determined by multiplying the normalized value of the parameter by its specified weight. Equation (3) shows the calculation of the total subindex of the $i$th parameter ($\text{Total}\_\text{S}{\text{I}}_i$): $$\begin{array}{}\text{(3)}& \text{Total}\_\text{S}{\text{I}}_i=W_i\ast N_i\text{pH}+\text{DO}+\text{COD}+\text{BOD}+\text{TC}.\end{array}$$

The WQI is calculated by adding together all the different indices for the sample shown in Equation (4): $$\begin{array}{}\text{(4)}& {\text{WQI}}_i=\frac{\text{Tota}{{\text{l}}_{\text{SI}}}_i}{\sum W_i}\ast 100.\end{array}$$

3.6. Correlation Analysis

Correlation analysis facilitates the identification of connections between various metrics of water quality. This information is vital to comprehend the intricate interaction among many factors that impact water quality. Figure 6 shows the heat map that illustrates the correlation between several characteristics.

[figure(s) omitted; refer to PDF]

The heat map plot depicting the correlation between several features reveals a robust link between the following pairs of features:

I. pH and TC

II. DO and BOD

III. COD and TC

The correlation coefficient between pH and TC is 0.8, indicating a robust positive association. As the pH level rises, there is a corresponding tendency for the TC to increase. The correlation coefficient between DO and BOD is −0.79, indicating a robust negative association. As DO levels grow, BOD generally decreases. This occurs because microbes utilize DO during the decomposition of organic matter in the BOD. The correlation coefficient between COD and TC is 0.89, indicating a robust positive correlation. As the cost of debt increases, there is a corresponding tendency for the TC to grow. The remaining pairs of features have weaker correlations. The correlation coefficient between pH and DO is 0.081, which indicates a weak positive correlation. The correlation coefficient between pH and COD is −0.0011, essentially zero. The correlation coefficient between BOD and DO is −0.75, which indicates a moderate negative correlation.

However, the heat map plot indicates that pH, DO, BOD, and TC are highly connected.

3.7. XGBoost

XGBoost is a robust machine learning algorithm with great potential. It has proven to be accurate in predicting water quality for applications on different scales, from small wells to large reservoirs and estuaries. It offers several advantages compared to traditional methodologies: (a) Water quality indicators include DO, pH value, and TC bacteria. With XGBoost, it is possible to accurately predict their values. (b) Its precision makes it a beneficial monitoring and control indicator for water quality. (c) XGBoost is very flexible, and its powerful ability to quickly handle a broad range of water quality data (categorical and numeric attributes) not only makes it a precious tool for forecasting the spread of blue–green algae but also puts up several appearances on screen. For these reasons, it is adaptable and can easily be tailored to fit many aquatic environments and various surveillance activities. (d) The XGBoost algorithm suits water quality data at different levels, spaces, and times. Due to its ability to process large datasets, it represents a new step forward in the analysis world. (e) Accordingly, the means for XGBoost’s feature importance scores can determine which attribute exerts the most significant influence on water quality within each location where the analysis takes place. Using these data in this way, more concrete water quality control programs can be specified.

This article suggests a simple and effective way to forecast water quality and use. This investigation used the methodology shown in Figure 7. Accuracy, precision, recall, $F1$ score, and confusion matrix are metrics for testing models predicting water quality. The systematic methodology of the research then examines the performance of a machine learning model based on a real-time dataset. We divided our dataset into 80% for training and 20% for testing.

[figure(s) omitted; refer to PDF]

3.8. Performance Evaluation/Validation of the Model

An evaluation was carried out to evaluate the efficacy of machine learning models for monitoring and prediction of water quality. The assessment employed measures generated from confusion matrices to analyze the performance of the models. The water quality classification used crucial evaluation metrics such as accuracy, precision, recall, and the $F1$ score in Equations (5), (6), and (7). The metrics expressed are computed using the following formulas: $$\begin{array}{}\text{(5)}& \text{Accuracy}A=\frac{\text{TP}+\text{TN}}{\text{TP}+\text{TN}+\text{FP}+\text{FN}}\text{(6)}& \text{Precision}P=\frac{\text{TP}}{\text{TP}+\text{FP}}\text{(7)}& \text{Recall}R=\frac{\text{TP}}{\text{TP}+\text{FN}}\end{array}$$

where TP, TN represent true positive and negative and FP, FN represent false positive and negative, respectively.

4. Experimental Results, Analysis, and Implications

During the data collection period of January 2022 to December 2022, 10 monitoring sites were chosen to evaluate the water quality of the Mahanadi River in various Odisha regions. The dataset was compiled according to season and belt type (including industrial, agricultural, and residential belts) using IoT sensors. The methodology outlined in Section 3.7 was used to calculate the WQI. The classification of water quality was determined by giving values to many characteristics of the water, as shown in Table 2.

Table 2

Categories of WQI measurements.

Sl. no.	WQI range	Category	Assigned value
1	0–50	Excellent	0
2	50–100	Good	1
3	100–200	Poor	2
4	200–300	Very poor	3

4.1. WQI for Months

Our research is aimed at examining water quality fluctuations every month using the WQI throughout the year. To determine the monthly WQI, the median value of the quality index assessments for the month is calculated.

Figure 8 indicates that the highest WQI (poorest water quality) is observed in summer, while the lowest WQI (best water quality) occurs in winter. This seasonal variation is primarily influenced by increased microbial activity, organic matter decomposition, and agricultural runoff during the warmer months. Bacterial proliferation and the release of organic substances from detritus can increase during the summer, which can contribute to the degradation of water quality. Monsoon season runoff from agricultural regions can be significantly increased due to the transport of contaminants into water bodies facilitated by heavy precipitation. In autumn, leaf decomposition can occur in bodies of water containing fallen trees. This process can lead to oxygen depletion and increased levels of BOD. In winter, lower temperatures can impede the rate of bacterial decomposition, resulting in the accumulation of organic waste in aquatic environments.

[figure(s) omitted; refer to PDF]

The WQI of the industrial belt is the highest, followed by the residential and agricultural belts as shown in Figure 9.

[figure(s) omitted; refer to PDF]

The line graph depicts the water quality characteristics measured at various stations over a year, with each station represented by a different line, and each month is shown in Figure 10.

I. pH: The pH values at all sites exhibit consistent stability throughout the year, ranging from 6.5 to 8.5. This falls within the ideal range for supporting aquatic organisms.

II. DO: The DO values of all sites fluctuate throughout the year, between 0.5 and 9 mg/L. In some sites, the DO level is not suitable for aquatic and drinkable purposes.

III. BOD: BOD levels peak during May and June and are at their lowest during November and December across all stations. The probable cause is the increased discharge of organic materials from agricultural activities during the monsoon period.

IV. TC: The TC peaks in May and June, while it is at its lowest in November and December for all stations. The probable cause is the increased discharge of fecal coliforms resulting from agricultural activities and human waste during the monsoon period.

V. COD: The concentration of COD peaks during May and June, reaching its lowest levels in November and December for all stations. The probable cause is the increased discharge of organic matter from agricultural activities and human waste during the monsoon period.

[figure(s) omitted; refer to PDF]

There is variation between the stations with respect to the water quality parameters, and certain stations demonstrate exceptional conditions. Compared to Kathajodi and Sankhatras stations, the water quality at Paradeep and Jharsugoda consistently improves. The water quality metrics at all stations demonstrate a seasonal pattern in which the monsoon season provides the most unfavorable conditions. Increased levels of COD, TCs, and BOD are commonly ascribed to the increased discharge of agricultural runoff and human waste during the monsoon season. Except for the elevated levels of DO, BOD, TC, and COD recorded during the monsoon season, the water quality indicators at all locations are adequate.

The categorical plots of several water quality metrics are shown in Figure 11. The excellent water quality has a narrower pH range than the good water quality. As a result, pH is a more sensitive indication of water quality than the other measures. DO values corresponding to exceptional water quality are more significant than those of good water quality. This is because DO is required for aquatic life to survive. Excellent water quality is characterized by lower COD and BOD ranges than acceptable water quality. This finding indicates that water quality is positively correlated with minimal levels of organic matter. The TC range for outstanding water quality is lower than the TC range for medium water quality. This indicates that minimal bacterial concentrations signify water quality.

[figure(s) omitted; refer to PDF]

The graph indicates that the pH, DO, COD, BOD, and TC parameters are critical water quality indicators. By monitoring these factors, we can better understand the general health of a water body and adopt protective measures.

4.2. Results for Water Quality Prediction

Machine learning models were used to forecast water quality categories, specifically “excellent,” “good,” and “poor.” The results of the experiments conducted to estimate water quality using machine learning models are provided in Table 3. The findings are relevant for the forecasting of water quality.

Table 3

Experimental result of the XGBoost ML model.

Metrics	Accuracy	Precision	Recall	$\mathit{F}\mathbf{1}$ score
Values	0.95	0.96	0.95	0.95

The machine learning models show excellent overall performance, as all measures surpass a threshold of 0.95. This suggests that the models can make correct predictions regarding water quality categories, exhibiting a significant level of precision and recall. The $F1$ score, a metric that calculates the harmonic mean of precision and recall, provides evidence of the models’ balanced performance.

➢ The accuracy value obtained of 0.95 indicates that the models successfully classified 95% of events related to water quality.

➢ With a precision value of 0.96, it can be identified that 96% of the instances categorized as “excellent,” “good,” or “poor” were classified accurately.

➢ The recall value of 0.95 indicates that the models accurately classified 95% of the water quality instances as “excellent,” “good,” or “poor.”

➢ The $F1$ score of 0.95 indicates that the models demonstrate a wonderful balance between recall and precision.

The confusion matrix in Figure 12 indicates that the model exhibits high accuracy in predicting water quality as excellent and good. In contrast, its performance in predicting poor water quality is comparatively lower. This is because there is a higher occurrence of false negatives, where the model incorrectly predicts good water quality despite the natural water quality being poor, compared to false positives, where the model incorrectly predicts poor water quality despite the actual water quality being good.

[figure(s) omitted; refer to PDF]

However, the model exhibits high precision and recall rates for excellent and good water quality but lower rates for bad water quality. These findings indicate that the model may exhibit a bias towards accurately forecasting high-quality water conditions, leading to the omission of instances of low-quality water conditions.

During our model implementation, we observed a training accuracy of 100% after 20 epochs and the test accuracy of our model reaches 95% after 20 epochs and this trend is almost constant up to 50 epochs, as shown in Figure 13. This is a positive indication since it suggests that the model is learning from the data and improving its ability to forecast water quality categories. The learning curve indicates that the model can accurately forecast water quality categories.

[figure(s) omitted; refer to PDF]

However, the findings of this analysis indicate that XGBoost has the potential to be a highly effective method for forecasting water quality classifications.

Figure 14 illustrates the accuracy of the XGBoost model in predicting WQI across different monitoring stations. The solid line represents experimental WQI values, while the dashed line represents model-predicted WQI. The close alignment between the two indicates the model’s reliability in assessing water quality trends.

[figure(s) omitted; refer to PDF]

4.3. Comprehensive Analysis

To show the importance of our proposed approach, we compared performance in this study. In the Related Work section, several carefully selected studies were analyzed that address the current problem. Researchers used IoT sensors and statistical techniques to collect online data throughout the experiment [35]. An alternative method was [36] the use of long- and short-term memory (LSTM) to forecast water quality. A summary of these studies is provided in Table 4, which shows that the proposed methodology unambiguously outperformed existing studies in all scenarios.

Table 4

Comparison analysis of recent work with the proposed work.

Works	River name	Data frame	IoT used	WQI calculated	Seasonal	Belts	Statistical/ML/DL model
[35]	Citarum River	2018–2020	√	X	X	X	√
[36]	Yangtze River	2016–2018	√	X	X	X	√
[40]	Weihe River	—	√	X	X	X	√
[37]	Satluj River	—	X	√	X	X	√
[41]	Thamirabarani River	May 2021–Dec 2022	X	X	X	X	√
[1]	Sangam (Ganga and Yamuna River)	Oct 2019–Feb 2020	√	√	√	X	X
[38]	Mahanadi River	2011–2019	X	√	X	X	√
[39]	Mahanadi River	2000–2018	X	X	X	X	√
Proposed work	Mahanadi River	January–December 2022	√	√	√	√	√

Approaches to assign water quality monitoring responsibilities based on clustering of ant colonies algorithmic were presented in [40]. Forecasting the quality of the water of the Satluj River was performed using various statistical methodologies [37]. Researchers using a hybrid deep learning model [41] predicted the water quality of the Thamirabarani River. The comparisons of WQI between seasons were derived from IoT sensors installed at the Sangam of Ganga and Yamuna to collect real-time data [1]. The Mahanadi River was investigated statistically and with artificial neural networks (ANNs) [38]. The water quality of the Mahanadi basin was studied using a multivariate statistical method [39]. These approaches were used to analyze the dataset we studied, such as calculating WQI and seasonal analysis and belt-wise comparison of WQI. Then, their results were compared with the proposed methodology. An impressive 95% accurate prediction model was developed with XGBoost, a machine learning model.

4.4. Implications of the Investigation

The implications of our proposed research, which centers on the utilization of IoT sensors, WQI calculations, and predictive models to monitor and forecast water quality in the Mahanadi River, are of immediate significance to the agricultural, residential, and industrial requirements of the inhabitants of the Mahanadi belt. River water, as an essential component to support life on our planet, profoundly impacts the well-being of organisms. However, the degradation of river water quality has resulted from the accelerating global development. The increasing magnitude of the human population exacerbates the need for copious amounts of potable water. Contrary to expectations, the urbanization necessary to facilitate this increase in population negatively affects the quantity and quality of water. Within this framework, our research underscores the critical nature of collaborative endeavors to assess water quality around the world, thereby highlighting the considerable influence that machine learning techniques can exert in this field. The imminent predicament of water scarcity is a significant concern that presents a danger to many nations in the decades to come. In addition to its worldwide importance, our research has practical implications for the Mahanadi region as it tackles the day-to-day needs of industrial sectors, residential zones, and agriculture. By incorporating IoT sensors, predictive models, and WQI calculations, a comprehensive strategy is achieved to monitor and forecast water quality. Figure 3 illustrates the proposed architecture, defining a methodical framework suitable for practical implementations. A range of sensors, such as pH, DO, BOD, COD, and TC, are utilized to collect a variety of water parameters. Precise water quality evaluations are achieved after entering these values into the predictive model.

5. Conclusion and Future Work

Water quality must be monitored regularly to ensure human survival, as it is the most essential ingredient for life on Earth. For this research, a 10-site monitoring network was strategically positioned to conduct a comprehensive assessment of water quality along the Mahanadi River in different Odisha regions. From January to December 2022, the data collected by IoT sensors was classified into distinct seasons and belts (industrial, agricultural, and residential). The preprocessing of the massive dataset, consisting of 121 rows and 13 columns, was thorough and included data cleansing and min–max normalization. This normalization technique standardized characteristics and improved water quality assessment and forecast models. In order to facilitate a thorough evaluation, weights were assigned to particular parameters, subindices were developed, and the WQI was computed. Stations such as Jharsugoda and Paradeep generally have higher water quality than other areas, with pronounced seasonal fluctuations, most notably during the monsoon season. Significant discoveries are increased concentrations of COD, TC, and BOD during the monsoon, which can be attributed to increased human waste and agricultural effluent. The importance of pH was underscored by the fact that it was a sensitive indicator with a more limited range for exceptional water quality. The correlation between critical parameters (pH, DO, COD, BOD, and TC) and the overall health of the water bodies underscored their importance. Machine learning models, more precisely, XGBoost, were implemented to predict water quality categories with a notable 95% accuracy. A comprehensive evaluation of the proposed method on established approaches, including assessments of seasonal patterns, WQI computation, and belt-wise comparisons, revealed its superior performance. The precision of the predictive model provided additional evidence of its effectiveness, which established it as a potentially helpful instrument for evaluating water quality in the Mahanadi River. The findings of this study, which emphasize validation and forecasting while utilizing advanced machine learning models, highlight its potential to address the critical need for adequate water quality assessment and management, thereby providing valuable information for environmental conservation and sustainable water management practices.

Although the proposed IoT-enabled XGBoost-based water quality prediction system exhibits a commendable accuracy rate of 95%, it is not without its limitations.

i. For real-time monitoring to be accurate, sensor calibration must be done periodically to prevent data drift.

ii. The model is trained on 10 monitoring sites, restricting its applicability to other river systems with various pollution sources.

iii. The approach does not explicitly account for abrupt pollution events like industrial spills, which could affect real-time estimates.

XGBoost performs well, but its computational complexity may limit real-time adoption on resource-constrained IoT devices. Addressing these restrictions would improve the suggested approach’s reliability.

Future research should focus on improving sensor calibration and precision by incorporating self-calibrating sensors or developing automated recalibration mechanisms to mitigate data drift. Applying the model to numerous river basins with different ecological and industrial situations improves its scalability and generalization. Finally, merging IoT sensor data with satellite imaging, weather data, and remote sensing technologies can improve water quality management prediction and early warning. These enhancements will make river ecosystem water quality monitoring more robust, scalable, and flexible.

Funding

The authors declare that no funding was received for this research.

Glossary

Nomenclature

IoTInternet of Things

MLmachine learning

GSMglobal system for mobile communications

WQIwater quality index

DOdissolved oxygen

BODbiochemical oxygen demand

CODchemical oxygen demand

TCstotal coliforms

RWMSriver water monitoring system

SIsubindex

$W_i$weights

$N_i$normalized values

TPtrue positive

TNtrue negative

FPfalse positive

FNfalse negative

References

[1] M. Kumar, T. Singh, M. K. Maurya, A. Shivhare, A. Raut, P. K. Singh, "Quality Assessment and Monitoring of River Water Using IoT Infrastructure," IEEE Internet of Things Journal, vol. 10 no. 12, pp. 10280-10290, DOI: 10.1109/JIOT.2023.3238123, 2023.

[2] L. Wang, X. Han, Y. Zhang, Q. Zhang, X. Wan, T. Liang, J. Rinklebe, "Impacts of Land Uses on Spatio-Temporal Variations of Seasonal Water Quality in a Regulated River Basin, Huai River, China," Science of The Total Environment, vol. 857,DOI: 10.1016/j.scitotenv.2022.159584, 2023.

[3] V. P. Dheeraj, C. S. Singh, A. K. Sonkar, N. Kishore, "Assessment of Surface and Ground Water Quality Around Korba Coalfield, India: An Integrated Approach of Water Quality Index, Multivariate Statistics Analysis and GIS Technique," Sustainable Water Resources Management, vol. 9 no. 6,DOI: 10.1007/s40899-023-00973-5, 2023.

[4] M. G. Uddin, S. Nash, A. Rahman, A. I. Olbert, "A Sophisticated Model for Rating Water Quality," Science of the Total Environment, vol. 868,DOI: 10.1016/j.scitotenv.2023.161614, 2023.

[5] R. Indugula, H. Venkata Vajjha, "Spatial Variations Of Trace Metals, Salinization Processes and Multivariate Analysis in the Near Surface Aquifer Systems of Krishna Delta, Andhra Pradesh, India," Environmental Earth Sciences, vol. 82 no. 10,DOI: 10.1007/s12665-023-10910-6, 2023.

[6] V. P. Dheeraj, C. S. Singh, N. Kishore, A. K. Sonkar, "Groundwater Quality Assessment in Korba Coalfield Region, India: An Integrated Approach of GIS and Heavy Metal Pollution Index (HPI) Model," Nature Environment & Pollution Technology, vol. 22 no. 1, pp. 369-382, DOI: 10.46488/NEPT.2023.v22i01.036, 2023.

[7] S. R. Laha, B. K. Pattanayak, S. Pattnaik, "Advancement of Environmental Monitoring System Using IoT and Sensor: A Comprehensive Analysis," AIMS Environmental Science, vol. 9 no. 6, pp. 771-800, DOI: 10.3934/environsci.2022044, 2022.

[8] S. R. Laha, B. K. Pattanayak, S. Pattnaik, D. Mishra, D. S. K. Nayak, B. B. Dash, "An IOT-Based Soil Moisture Management System for Precision Agriculture: Real-Time Monitoring and Automated Irrigation Control," Proceedings pf the 2023 4th International Conference on Smart Electronics and Communication (ICOSEC), pp. 451-455, DOI: 10.1109/ICOSEC58147.2023.10276266, 2023.

[9] A. J. Ramadhan, A. M. Ali, H. K. Kareem, "Smart Water-Quality Monitoring System Based on Enabled Real-Time Internet of Things," Journal of Engineering Science and Technology, vol. 15 no. 6, pp. 3514-3527, 2020.

[10] M. S. U. Chowdury, T. B. Emran, S. Ghosh, A. Pathak, M. M. Alam, N. Absar, M. S. Hossain, "IoT Based Real-Time River Water Quality Monitoring System," Procedia Computer Science, vol. 155, pp. 161-168, DOI: 10.1016/j.procs.2019.08.025, 2019.

[11] Y. Li, X. Wang, Z. Zhao, S. Han, Z. Liu, "Lagoon Water Quality Monitoring Based on Digital Image Analysis and Machine Learning Estimators," Water Research, vol. 172,DOI: 10.1016/j.watres.2020.115471, 2020.

[12] A. G. Aliyu, N. R. B. Jamil, M. B. B. Adam, Z. Zulkeflee, "Spatial and Seasonal Changes in Monitoring Water Quality of Savanna River System," Arabian Journal of Geosciences, vol. 13 no. 2,DOI: 10.1007/s12517-019-5026-4, 2020.

[13] L. Cui, M. Qiu, H. Fan, L. Wang, S. Zhao, J. Du, "Design and Research of Water Quality Monitoring System Based on nRF24L01 Wireless Communication," IOP Conference Series: Earth and Environmental Science, vol. 208 no. 1,DOI: 10.1088/1755-1315/208/1/012074, 2018.

[14] N. M. Ragi, R. Holla, G. Manju, "Predicting Water Quality Parameters Using Machine Learning," Proceedings of the 2019 4th International Conference on Recent Trends on Electronics, Information, Communication & Technology (RTEICT), pp. 1109-1112, DOI: 10.1109/RTEICT46194.2019.9016825, 2019.

[15] R. P. N. Budiarti, S. Sukaridhoto, M. Hariadi, M. H. Purnomo, "Big Data Technologies using SVM (Case Study: Surface Water Classification on Regional Water Utility Company in Surabaya)," Proceedings of the 2019 International Conference on Computer Science, Information Technology, and Electrical Engineering (ICOMITEE), pp. 94-101, DOI: 10.1109/ICOMITEE.2019.8920823, 2019.

[16] R. S. Sharmila, R. Sushma, M. U. Mahanth, A. Chandana, C. Sunil, "Water Pollution Monitoring System Using IoT," ICDSMLA 2020. Lecture Notes in Electrical Engineering, vol. 783,DOI: 10.1007/978-981-16-3690-5_111, 2022.

[17] A. Porwal, S. K. Mishra, A. Kela, G. Singh, M. Panchal, A. Jain, "Prediction and Prevention of Water Pollution by Industries using IoT based Monitoring System," Proceedings of the 2022 IEEE 11th International Conference on Communication Systems and Network Technologies (CSNT), pp. 504-509, DOI: 10.1109/CSNT54456.2022.9787664, 2022.

[18] J. Buhion, J. Buhion, J. Ocon, "The Development of a Real Time, Interactive Water Level Monitoring System Through SMS With AC Load Control," Technium, vol. 4 no. 5, pp. 67-76, DOI: 10.47577/technium.v4i5.6698, 2022.

[19] K. Saravanan, E. Anusuya, R. Kumar, L. H. Son, "Real-Time Water Quality Monitoring Using Internet of Things in SCADA," Environmental Monitoring and Assessment, vol. 190 no. 9,DOI: 10.1007/s10661-018-6914-x, 2018.

[20] G. George, N. N. Menon, A. Abdulaziz, R. J. Brewin, P. Pranav, A. Gopalakrishnan, T. Platt, "Citizen Scientists Contribute to Real-Time Monitoring of Lake Water Quality Using 3D Printed Mini Secchi Disks," Frontiers in Water, vol. 3,DOI: 10.3389/frwa.2021.662142, 2021.

[21] Z. Wu, J. Wang, C. Bian, J. Tong, S. Xia, "A MEMS-Based Multi-Parameter Integrated Chip and Its Portable System for Water Quality Detection," Micromachines, vol. 11 no. 1,DOI: 10.3390/mi11010063, 2020.

[22] R. Schima, S. Krüger, J. Bumberger, M. Paschen, P. Dietrich, T. Goblirsch, "Mobile Monitoring—Open-Source Based Optical Sensor System for Service-Oriented Turbidity and Dissolved Organic Matter Monitoring," Frontiers in Earth Science, vol. 7,DOI: 10.3389/feart.2019.00184, 2019.

[23] A. U. Alam, D. Clyne, W. Lush, M. J. Deen, "A Reusable, Reagent-Less Free Chlorine Sensor Using Gold Thin Film Electrode," Analyst, vol. 146 no. 8, pp. 2626-2631, DOI: 10.1039/D1AN00038A, 2021.

[24] G. Matta, A. Kumar, A. Nayak, P. Kumar, A. Kumar, P. K. Naik, S. K. Singh, "Assessing Heavy Metal Index Referencing Health Risk in Ganga River System," International Journal of River Basin Management, vol. 21 no. 4, pp. 759-769, DOI: 10.1080/15715124.2022.2098756, 2023.

[25] R. Martínez, N. Vela, A. El Aatik, E. Murray, P. Roche, J. M. Navarro, "On the Use of an IoT Integrated System for Water Quality Monitoring and Management in Wastewater Treatment Plants," Water, vol. 12 no. 4,DOI: 10.3390/w12041096, 2020.

[26] A. T. Constantin, G. I. Lazar, S. V. Nicoara, "Water Quality Monitoring Sensors Optimum Location in a Metropolitan Supply Network," Revista de Chimie, vol. 71 no. 1, pp. 327-334, DOI: 10.37358/RC.20.1.7853, 2020.

[27] M. Zhu, J. Wang, X. Yang, Y. Zhang, L. Zhang, H. Ren, L. Ye, "A Review of the Application of Machine Learning in Water Quality Evaluation," Eco-Environment & Health, vol. 1 no. 2, pp. 107-116, DOI: 10.1016/j.eehl.2022.06.001, 2022.

[28] M. Ghashghaie, H. Eslami, K. Ostad-Ali-Askari, "Applications of Time Series Analysis to Investigate Components of Madiyan-Rood River Water Quality," Applied Water Science, vol. 12 no. 8,DOI: 10.1007/s13201-022-01693-5, 2022.

[29] H. Xu, Q. Gao, B. Yuan, "Analysis and Identification of Pollution Sources of Comprehensive River Water Quality: Evidence From Two River Basins in China," Ecological Indicators, vol. 135,DOI: 10.1016/j.ecolind.2022.108561, 2022.

[30] E. Gençer, C. Başaran, "Water Quality Assessment and Pollution of Akarçay River, Türkiye," Kuwait Journal of Science, vol. 51 no. 1,DOI: 10.1016/j.kjs.2023.07.003, 2024.

[31] H. S. Cadraku, A. Beqiraj, "Assessment of Water Quality in Blinaja River Basin (Kosovo) Using the Canadian Water Quality Index (WQI)," Journal of Survey in Fisheries Sciences, vol. 10 no. 1, pp. 29-47, 2023.

[32] R. Kkhalili, M. Parvinnia, A. Zali, "Water Quality Assessment of Garmarood River Using the National Sanitation Foundation Water Quality Index (NSFWQI), River Pollution Index (RPI) and Weighted Arithmetic Water Quality Index (WAWQI)," Environment and Water Engineering, vol. 6 no. 3, pp. 273-283, DOI: 10.22034/jewe.2020.238090.1381, 2020.

[33] M. G. Uddin, S. Nash, A. Rahman, A. I. Olbert, "A Comprehensive Method for Improvement of Water Quality Index (WQI) Models for Coastal Water Quality Assessment," Water Research, vol. 219,DOI: 10.1016/j.watres.2022.118532, 2022.

[34] B. Dutta, B. Sarma, "Assessment of Water Quality Index of the Kolong River of Nagaon District of Assam, India," Journal of Engineering Research and Application, vol. 8 no. 6, pp. 29-38, 2018.

[35] A. B. Pantjawati, R. D. Purnomo, B. U. D. I. Mulyanti, L. A. Z. U. A. R. D. I. Fenjano, R. E. Pawinanto, A. B. D. Nandiyanto, "Water Quality Monitoring in Citarum River (Indonesia) Using IoT (Internet of Thing)," Journal of Engineering Science and Technology, vol. 15 no. 6, pp. 3661-3672, 2020.

[36] P. Liu, J. Wang, A. K. Sangaiah, Y. Xie, X. Yin, "Analysis and Prediction of Water Quality Using LSTM Deep Neural Networks in IoT Environment," Sustainability, vol. 11 no. 7, 2019.

[37] S. Tiwari, R. Babbar, G. Kaur, "Performance Evaluation of Two ANFIS Models for Predicting Water Quality Index of River Satluj (India)," Advances in Civil Engineering, vol. 2018 no. 1,DOI: 10.1155/2018/8971079, 2018.

[38] R. Pany, A. Rath, P. C. Swain, "Water Quality Assessment for River Mahanadi of Odisha, India Using Statistical Techniques and Artificial Neural Networks," Journal of Cleaner Production, vol. 417,DOI: 10.1016/j.jclepro.2023.137713, 2023.

[39] A. Das, "Multivariate Statistical Approach for the Assessment of Water Quality of Mahanadi Basin, Odisha," Materials Today: Proceedings, vol. 65, pp. A1-A11, DOI: 10.1016/j.matpr.2022.08.146, 2022.

[40] T. Dang, J. Liu, "Design of Water Quality Monitoring System in Shaanxi Section of Weihe River Basin Based on the Internet of Things," Computational Intelligence and Neuroscience, vol. 2022,DOI: 10.1155/2022/3543937, 2022.

[41] B. Perumal, N. Rajarethinam, A. D. Velusamy, V. P. Sundramurthy, "Water Quality Prediction Based on Hybrid Deep Learning Algorithm," Advances in Civil Engineering, vol. 2023,DOI: 10.1155/2023/6644681, 2023.