A Metaheuristic-Optimized Ensemble Learning Framework for Sustainable Water Quality Forecasting in Multi-Sensor Environmental Monitoring Systems
Abstract
Context: For long-term environmental management, it is important to accurately predict water quality, especially as modern multi-sensor monitoring systems generate large volumes of high-dimensional physicochemical data. Conventional statistical and machine learning methodologies frequently fail to capture nonlinear environmental dynamics or rely on manually adjusted model configurations, thereby limiting predictive accuracy and scalability. Objective: To tackle these issues, this research introduces a metaheuristic-optimized ensemble learning framework for long-term water quality prediction in multi-sensor environmental monitoring systems. Method: The proposed framework combines Particle Swarm Optimization (PSO) with Random Forest ensemble learning to enable automatic hyperparameter optimization within a repeatable end-to-end modeling pipeline. Comprehensive data preprocessing and domain-informed feature engineering were applied to better represent physicochemical interactions and temporal environmental patterns. Results: The experiment was conducted using a large-scale water quality dataset comprising 24,265 validated samples. Conclusion: The framework provides a metaheuristic-optimized ensemble learning approach for long-term water quality prediction in multi-sensor environmental monitoring systems.
