Forecasting Quality of Water Using Leading Machine Learning Strategy

Degradation in the quality of water is one of the greatest challenges facing the world at large, which greatly affects human health and safety, agriculture, and ecosystem health. Those industrial and agricultural, as well as household waste pollutants, still threaten the safety of drinking water and therefore need urgent attention from large-scale efficient monitoring solutions. Previous approaches of water quality assessment are labor-intensive and more time is needed, where an increasing demand has developed for automatic and accurate predictive models to report in real-time. To meet this challenge, the proposed paper introduces a machine learning-based approach toward predicting water quality that falls into three distinct classes: poor, moderate, or good. The paper analyzes the predictive capability of five models: TabNet, KNNeighbour, CatBoost, LightGBM, and Decision Tree as well, showing a consistent trend in improvement in prediction precision. These models performances were measured considering MSE and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{2}$</tex> scores for validation of the reliability of the predictions. A proposed solution is aimed at improving the management of water resources and at providing actionable information to decision-makers to mitigate pollution and safeguard public health and ecological systems. The top-performing model, LightGBM, had an accuracy of 87.04%, an RMSE of 0.36, and an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{2}$</tex> score of 0.26, which indicates the superior predictability performance for water quality estimation.