Modeling and prediction of river quality parameters based on a novel hybrid machine learning model

Abstract

Water pollution in rivers is a major issue, and modeling water quality using machine learning techniques is an effective approach. Total dissolved solids is a key water quality indicator, as improper levels can affect industrial, agricultural, and urban water use. Least Squares Support Vector Machine (LSSVM) algorithms are highly accurate for modeling and predicting water quality parameters, especially when combined with the Crayfish Optimization Algorithm (COA) to enhance prediction accuracy. A study using data from the Jajrood basin (Latiyan, Rodak, and Sharifabad stations) applied LSSVM and LSSVM-COA to model and predict key water quality parameters, including CO₃⁻², HCO₃⁻¹, Cl⁻¹, So₄⁻², Ca⁺², Mg⁺², Na⁺¹ and K⁺¹. The accuracy of the results varied, but the LSSVM-COA algorithm was more accurate, particularly at Sharifabad station, where the R values for the combined parameters during testing were 0.92 and the RRMSE was 0.40. Predictive analysis indicated that parameters like Cl⁻¹, So₄⁻² and Ca⁺² had the most significant impact on increasing TDS, especially when each parameter's value was increased by 50 %.