Development of an Intelligent Tool for Water Quality Assessment

Abstract

Water quality monitoring is essential for managing and protecting surface water resources. This study investigates the assessment and prediction of water quality using both analytical and machine learning approaches. Analytical methods for calculating the Water Quality Index (WQI) aggregate multiple water quality parameters into a single value representing overall water quality. Adopting reliable machine learning methods for predicting the WQI is crucial for achieving accurate results by leveraging computational processing to approximate the WQI from complex, combined input variables. This study evaluates the efficacy of Support Vector Regression (SVR) and Random Forest Regression (RFR) in predicting the WQI of surface water. The WQI was calculated using the Weighted Arithmetic Method to establish baseline water quality classes and assess the surface water quality at Tilesdit Dam in Bouira, Algeria, based on a dataset of six key physicochemical parameters: pH, temperature, electrical conductivity, turbidity, total dissolved solids, and dissolved oxygen. The performance of the prediction models was evaluated using multiple statistical metrics. Performance evaluation demonstrated that SVR provided superior accuracy, with R² values exceeding 0.999 and minimal prediction errors. Feature importance analysis using the RFR model indicated that turbidity was the most influential parameter for accurate WQI prediction. SVR successfully predicted the WQI with high accuracy. Although RFR yielded lower prediction results than SVR, it remains valuable for feature selection and identifying the most relevant variables. Within this dual-task framework, RFR determines variable importance, while SVR performs the final WQI prediction. The proposed framework enables intelligent, real-time water quality monitoring with reduced reliance on laboratory analyses, providing a cost-effective, rapid, and reliable tool for environmental decision-making and management.