Developing river water quality prediction model incorporating reliable indexing approach

Hyperparameters optimization techniques can influence the prediction model(s) capabilities in terms of acquiring reliable prediction results. As a part of the development of the data-driven water quality (WQ) model(s), this research evaluated five hyperparameters optimization techniques and their impacts on WQ prediction model(s) using machine learning (ML)/artificial intelligence (AI) techniques to predict the WQI scores by obtaining the root mean squared (RMS)-WQI technique. For developing the ML/AI models the research utilized the ten ML algorithms by comparing fifty models. To evaluate the ML-AI model(s), the study used five widely used metrics including Root Mean Squared Error (RMSE), Mean Squared Error (MSE), Mean Absolute Error (MAE), Percentage of Absolute Bias Error (PABE), Nash Sutcliffe Efficiency (NSE) while coefficient of determination (R2) utilized for sensitivity assessment. The results indicated that the gradient boosting (GBR) model with the OPTUNA (OPT) optimization technique outperformed in predicting WQI score during both training (RMSE = 0.84, MSE = 0.71, MAE = 0.74, PABE = 1.00) and testing (RMSE = 0.45, MSE = 0.20, MAE = 0.30, PABE = 0.41) phase, compared to the other prediction models. Additionally, the study also revealed that the GBR-OPT demonstrated higher sensitivity (R2 for the year 2021 = 0.99 and testing R2 for the year 2022 = 0.98) and the highest efficient (an average NSE for the year 2021 = 0.71 and an average NSE for the year 2022 = 0.60) compared to others. However, the findings of the research reveal that the study outcomes could be effective in developing the more efficient and accurate WQ prediction model(s) that would be helpful for sustainable WQ management.