Optimizing Parameter Selection for Forecasting and Classifying Water Quality

Abstract

Abstract Effective water quality monitoring systems are essential for environmental management and public health; however, traditional monitoring methods can be inefficient, costly, and resource-intensive. This work examined India and Hong Kong as case studies, focusing on optimizing water quality parameters for improved system performance while reducing costs and power consumption. Through preprocessing techniques, including correlation matrix analysis, sensitivity analysis, and ANOVA, the initial set of seven water quality parameters was reduced to four: DO, pH, BOD, and Fecal Coliform. To overcome limitations in conventional assessment methods, deep learning models were employed to enhance water quality forecasting and classification. LSTM was used to forecast WQI values, achieving a RMSE of 0.2498. Meanwhile, CNN classified water with 97.20% accuracy, while also forecasting the selected water quality parameters. This work contributes to the field by demonstrating a lightweight, low-power deep learning approach that achieves comparable or superior performance using fewer parameters, making it suitable for real-time water monitoring in resource-limited environments.