Predicting and investigating water quality index by robust machine learning methods.
Journal:
Journal of environmental management
PMID:
40179471
Abstract
This study addresses the critical challenges of waste management and water quality in urban environments, where accelerated urbanization has exacerbated environmental degradation and public health risks. Employing advanced machine learning algorithms-Long Short-Term Memory (LSTM), Random Forest (RF), Decision Tree (DT), and Support Vector Machine (SVM)-this research predicts the Water Quality Index (WQI) to improve urban environmental management. The novelty lies in the integration of multiple algorithms within a single framework, focusing on classifying WQI labels (1-9) for "good" to "poor" water quality, a departure from traditional continuous value predictions. Among the algorithms, LSTM demonstrated the most significant advantages, achieving superior predictive accuracy and precision across training, testing, and validation datasets, with RMSE values of 0.0611, 0.0810, and 0.0754 and R values consistently above 0.9964. Comparative analysis revealed LSTM's capacity to capture complex temporal dependencies in data, surpassing RF, DT, and SVM in predictive performance. This approach provides actionable insights into WQI dynamics, enabling the identification of key pollution factors, optimizing waste management practices, and supporting real-time decision-making. The integration of climate indicators into the models further enhances their applicability in addressing long-term trends associated with climate change. Statistical evaluations, including AARE and SD, corroborate LSTM's robustness and reliability, making it a transformative tool for urban water quality prediction. This research pioneers a scalable, efficient, and practical solution to urban environmental challenges, contributing to sustainable resource management and improved public health outcomes.