EXPLAINABLE ENSEMBLE MACHINE LEARNING FRAMEWORK FOR WATER QUALITY ASSESSMENT USING PHYSICOCHEMICAL INDICATORS

Abstract

Water quality monitoring is one of crucial activity for sustainable water resource management and global environmental protection. The conventional assessment techniques mostly rely on the threshold value based analysis of physicochemical data. Such methods may fail to capture complex and nonlinear relationships in the water quality parameters. In our study, a machine learning–based framework has proposed for classification of water quality by using physicochemical parameters. The parameters in dataset includes pH, hardness, solids, conductivity, organic carbon, and turbidity. The proposed methodology for this work includes an enhance data preprocessing strategy for imputation of adaptive missing-values using Mean, KNN and Iterative (MICE). Then evaluation of multiple ML models, logistic regression, support vector machines, random forests, and gradient boosting were performed for optimization of model. The stacking ensemble model combining heterogeneous base learners is developed for the enhancement of model classification and its performance. The efficiency of Model is assessed on evaluation metrics, including accuracies, precision, recall, F1-score, and ROC curves. The results demonstrated an enhanced performance for ensemble-based models compared to individual classifiers. Moreover, explainable artificial intelligence based on Shapley Additive Explanations, known as SHAP, have been adopted to interpret model’s prediction. The results show the effectiveness of ensemble machine learning and explainable AI for robust and interpretable water quality assessment. This can be useful offering for a data-driven based decision-support framework for environmental monitoring applications.