Machine Learning for Diagnosing Water Potability and Explainable AI for Contextual Insights

Abstract

Availability of water is one of the most important aspects of Earth’s status as the only planet capable of supporting life. Although water makes up 70% of the earth’s surface, the availability of drinkable water is extremely limited. Water makes up about 70% of the human body and aids in the healthy functioning of the human body. Contaminated water can have a pernicious effect on the human body, thus it’s important to find a safe drinking water source. Five machine learning algorithms were explored to estimate the potability of water in this study. Three regression algorithms are applied to estimate the missing values in this study. Among the implemented, a Deep Neural Network (DNN) model achieves a better accuracy of 66.1%, with precision, recall, and AUC scores of 61.2%, 35.8%, and 67%, respectively which is comparable with the present state-of-the-art. The Support Vector Machine (SVM) applied has achieved the highest precision and the lowest recall, despite having the second-highest accuracy of 65.1% in this study. AdaBoost (ADB) achieves the highest recall of 44.1%, as well as the highest AUC score of 74.5%. In addition, a local explanation artificial algorithm called LIME is applied to explain why a certain sample of water is potable.