A Machine Learning-Based Water Potability Prediction Model by Using Synthetic Minority Oversampling Technique and Explainable AI.

Abstract

During the last few decades, the quality of water has deteriorated significantly due to pollution and many other issues. As a consequence of this, there is a need for a model that can make accurate projections about water quality. This work shows the comparative analysis of different machine learning approaches like Support Vector Machine (SVM), Decision Tree (DT), Random Forest, Gradient Boost, and Ada Boost, used for the water quality classification. The model is trained on the Water Quality Index dataset available on Kaggle. Z-score is used to normalize the dataset before beginning the training process for the model. Because the given dataset is unbalanced, Synthetic Minority Oversampling Technique (SMOTE) is used to balance the dataset. Experiments results depict that Random Forest and Gradient Boost give the highest accuracy of 81%. One of the major issues with the machine learning model is lack of transparency which makes it impossible to evaluate the results of the model. To address this issue, explainable AI (XAI) is used which assists us in determining which features are the most important. Within the context of this investigation, Local Interpretable Model-agnostic Explanations (LIME) is utilized to ascertain the significance of the features.