Big Data Analytics in IoT, social media, NLP, and information security: trends, challenges, and applications

Abstract

This paper presents a comprehensive, domain-specific survey and experimental evaluation of machine learning techniques for Big Data Analytics across four critical domains: IoT, Social Media, Natural Language Processing (NLP), and Information Security. A novel taxonomic framework is introduced to classify and analyze the suitability of algorithms based on empirical, experimental, and computational perspectives. The study integrates large-scale experimental benchmarking of key techniques—including CNN, XGBoost, Self-Supervised Learning (SSL), Graph Neural Networks (GNN), ELM, KNN, and Decision Trees—using real-world and synthetic datasets, and evaluates them across five performance metrics: accuracy, F1-score, precision, recall, and computational time. Key findings reveal that: (1) GNN and Self-Supervised Learning (SSL) are top performers in terms of predictive performance and efficiency in domains such as IoT and Social Media, (2) XGBoost and CNN offer superior accuracy and robustness across structured and unstructured data tasks, though CNN incurs higher computational costs, (3) ELM and Decision Trees are better suited for lightweight or interpretable applications, and (4) KNN generally underperforms in scalability and predictive strength for large-scale tasks. The taxonomy and experiments collectively demonstrate the need for context-aware algorithm selection, particularly for real-time and scalable Big Data applications. By aligning algorithmic properties with domain-specific challenges, this study offers actionable insights for researchers and practitioners seeking effective analytic strategies in the evolving landscape of Big Data.