Enhancing Drinking Water Quality Assessment: An Exploration of Elemental Composition for Drinkable Water

Water is essential to human living; hence, maintaining its purity is vital to health. This work seeks to find the best algorithm for the statistically imputed dataset and the most accurate drinking water prediction algorithm. The datasets are used to estimate water potability using eight algorithms analyzing nine features. Our study uses hyperparameter tweaking and features engineering to increase the accuracy of numerous machine-learning models that predict drinkable water. PCA, ICA, and TSVD output are engineered to forecast water potability in diverse characteristics. Grid search is essential for optimizing and tweaking nine classification models’ hyperparameters. RF, Xgboost, LR, GNB, and Ada-Boost models are used to forecast water categorization. KNN, DT, SVR, and MLP models predict water portability. PCA, ICA, and TSVD have been used to indicate water potability. This study employed 9 characteristics and 3276 occurrences. Four metrics were calculated to evaluate categorization methods: accuracy, recall, precision, and score. This study explains predictions using Explain Like I’m Five (ELI5) and Shapash, two prominent Explainable AI (XAI) tools. The dataset was split into 80% training and 20% testing, with SMOTE, SMOTeTomek, and Near Miss approaches used for training balance. The XGB model had the highest accuracy (99.80%) in forecasting water potability, whereas the RF model had a higher accuracy (99.40%).