Prediction and Analysis of Water Quality Index using Machine Learning Approaches

Clean and reliable water is a vital factor for sustainability and public health. Traditional testing of water quality relies on manually selected tests, which take a lot of time and money. This study uses machine learning (ML) techniques to help automate and improve the prediction of the Water Quality Index (WQI). Factors are preliminary examination of the water and include physicochemical factors such as pH, chloramines, sulfate, conductivity, trihalomethanes, and turbidity. The data was collected from Kaggle and preprocessed using forward fill/backward fill, inject Gaussian noise (to augment the data), and removed outliers using the Interquartile Range method. For feature selection, a combination of Recursive Feature Elimination, and Random Forest algorithm were used, filling the role of a feature selection method. Three algorithms, KNN, LightGBM, and CatBoost were evaluated with R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> scores of 0.8835, 0.9295, and 0.9669, respectively. The CatBoost model showed better predictive performance with lower error metrics compared to both KNN and LightBG, and also provided accuracy in mapping both mean and median WQI scores. The proposed machine learning framework added value to water quality assessments for immediate decisions about water being tested, and about sustainability and time-saving methods.