Water quality prediction using Machine Learning Models

¹ Department of civil Engineering, Jaypee University of Information Technology Waknaghat, Solan Himachal Pradesh India, 173234.
² Department of Computer Science and Engineering, Himachal Pradesh University, Summer Hill, Shimla

^* Corresponding author: richasharma7372@gmail.com

Abstract

The quality of water is a vital determinant of environmental sustainability, economic development, and general welfare. India has substantial water quality issues, with different areas facing varying levels of pollution. Industrial effluents introduce toxic chemicals and heavy metals into water bodies, while agricultural runoff carries pesticides, fertilizers, and sediments, causing eutrophication and water pollution. The Ganges, Yamuna, and Godavari rivers have elevated amounts of pollution. According to the Central Pollution Control Board, the levels of biochemical oxygen demand, which is a measure of organic pollution, often above the acceptable thresholds in many sections of these rivers. Conventional techniques for monitoring water quality are often arduous, time-consuming, and incapable of delivering real- time evaluations. The objective of this study is to create a precise classification model that can accurately forecast water quality by using a range of indicators. The aim is to use machine learning techniques, including decision trees, K-Nearest Neighbor (KNN), and Random Forest, to develop prediction models that can effectively assess water quality and identify possible pollution incidents before they become major issues. This research used a comprehensive dataset of water quality metrics, including pH, turbidity, dissolved oxygen, temperature, phosphates, and nitrates, to assess the accuracy of each algorithm in forecasting water potability. The Random Forest method attained a superior accuracy of 70.4%, successfully handling intricate interactions and mitigating overfitting by using ensemble learning. The KNN method, which achieved an accuracy of 59%, had challenges arising from its susceptibility to the selection of k and distance measures, as well as processing inefficiencies. The Decision Tree approach, despite its speed and interpretability, had the lowest accuracy of 58% mostly owing to overfitting, which impeded its ability to generalize. This study highlights the better performance of the Random Forest model in predicting water quality because of its ability to capture complex non-linear relationships, handle noisy data, and prevent overfitting by aggregating multiple decision trees.