Machine Learning Approaches for Quality Analysis of Drinking Water

Water is an indispensable requirement for human life, but not all water is portable for human consumption. Therefore, the water quality (WQ) is assessed using various metrics, such as biological, chemical and physical metrics, regular laboratory testing and analysis which can be tiresome and time-consuming. Thus, advanced techniques such as machine learning, deep learning may be better option in terms of potability assessment. Several studies have been conducted on prediction of WQ keeping in view that multiple factors and classification can affect the prediction model. The quality of the forecasts depends mainly on the quality and relevance of the data used to train the model. Therefore, the dataset for this study is taken from the last six years of test data from National Accreditation Board for Testing and Calibration Laboratories (NABL) accredited / certified laboratories. This study investigates various supervised machine learning–based models such as Random Forest, Support Vector Machine, Decision Tree and Gradient Boosting Classifier, and unsupervised machine learning–based model K-Means, Bisecting K-Means and MiniBatch K-Means clustering algorithm with the ability to classify and evaluate the drinking water quality. Comparing the performance of mentioned models, it is revealed that supervised learning methods outperform the unsupervised clustering methods.