Real-Time Water Quality Forecasting Using a Simulation-Driven IoT and Machine Learning Framework for Resource-Constrained Smart Cities

This study proposes and evaluates a simulation-driven Internet of Things (IoT) and machine-learning framework for real-time water-quality monitoring and forecasting in resource-constrained smart cities. Ensuring timely and reliable water-quality data remains a critical challenge, as conventional monitoring approaches are often delayed, infrastructure-intensive, and difficult to scale. While IoT-based systems support continuous environmental monitoring, many existing implementations emphasise data acquisition rather than predictive analytics and deployment-oriented design. To address this gap, the study develops a unified simulation-based modelling framework in which virtual sensors generate key water-quality parameters, including pH, turbidity, temperature, and dissolved oxygen. The generated data are processed using a machine learning pipeline and used to train supervised models, namely XGBoost, Random Forest, and K-Nearest Neighbours. Model performance is evaluated using accuracy, precision, recall, and F1-score to ensure balanced assessment across water-quality conditions. The results show that XGBoost achieves the highest predictive performance (91% accuracy), demonstrating strong capability in modelling non-linear environmental patterns and detecting unsafe water conditions. In addition to predictive performance, the framework supports real-time forecasting and visualisation through an integrated dashboard, enabling continuous monitoring and data-driven decision-making. The study contributes a reproducible and scalable framework that integrates simulation-based modelling, IoT-enabled monitoring, and machine learning for predictive analytics. The proposed approach provides a cost-effective pathway for advancing sustainable water management and environmental monitoring in resource-constrained smart-city environments.