Assessing the feasibility of using Machine learning algorithms to determine reservoir water quality based on a reduced set of predictors

• Four ML models were analyzed for water quality prediction using the WQI index. • Analyses were performed for the entire (five water quality indicators) and reduced dataset. • All machine learning (ML) models demonstrated high accuracy and good predictive ability for the entire dataset. • The models achieved R 2 = 0.999 (NNM, LR) for five, 0.988 (KNN) for three and 0.941 (RF) for two predictors. • The application of ML algorithms enables rapid WQI and water quality prediction, even with a reduced set of indicators. The present study analyses the possibility of assessing water quality using the water quality index (WQI) through the application of four different machine learning algorithms (ML): neural network models (NNM), random forest (RF), k-nearest neighbor (KNN), and linear regression (LR). Water quality was determined based on 5 indicators: P, COD, BOD 5 , N total, and total suspended solids TS. The possibility of predicting water quality (WQI index) based on the reduced number of predictors was then analyzed. It was estimated which indicators have the most significant impact on WQI values. The performance of models using different algorithms, as well as those trained on full and reduced data sets, was compared. The models demonstrate high performance in WQI prediction, achieving an R 2 of 0.999 (for NNM and LR) for the entire dataset, 0.988 (KNN) for the dataset using only three types of predictors, and 0.941 for the dataset using only two predictors (RF). The construction and training of ML models for reduced sets and types of predictors will enable early water quality estimation based on only a few selected parameters. The implementation of ML algorithms will enable more effective water quality management and significantly improve the precision of predictions for critical water parameters.