Prediction of Water Quality Classification of the Kelantan River Basin, Malaysia, Using Machine Learning Techniques

Abstract

Machine Learning (ML) has been used for a long time and has gained wide attention over the last several years. It can handle a large amount of data and allow non-linear structures by using complex mathematical computations. However, traditional ML models do suffer some problems, such as high bias and overfitting. Therefore, this has resulted in the advancement and improvement of ML techniques, such as the bagging and boosting approach, to address these problems. This study explores a series of ML models to predict the water quality classification (WQC) in the Kelantan River using data from 2005 to 2020. The proposed methodology employed 13 physical and chemical parameters of water quality and 7 ML models that are Decision Tree, Artificial Neural Networks, K-Nearest Neighbors, Naïve Bayes, Support Vector Machine, Random Forest and Gradient Boosting. Based on the analysis, the ensemble model of Gradient Boosting with a learning rate of 0.1 exhibited the best prediction performance compared to the other algorithms. It had the highest accuracy (94.90%), sensitivity (80.00%) and f-measure (86.49%), with the lowest classification error. Total Suspended Solid (TSS) was the most significant variable for the Gradient Boosting (GB) model to predict WQC, followed by Ammoniacal Nitrogen (NH3N), Biochemical Oxygen Demand (BOD) and Chemical Oxygen Demand (COD). Based on the accurate water quality prediction, the results could help to improve the National Environmental Policy regarding water resources by continuously improving water quality.