Machine Learning‐Based Assessment of Groundwater Quality in Urban Odisha: A Case Study of Cuttack and Bhubaneswar

ABSTRACT Access to safe drinking water is essential for public health, yet rapid urbanization in cities like Cuttack and Bhubaneswar in India is exacerbating groundwater contamination. Industrial discharge and inadequate infrastructure are contributing to water quality deterioration, which calls for the development of robust frameworks for assessment and management. This study aims to predict the water quality index (WQI), classify groundwater quality, and identify key contaminants using machine learning (ML) techniques. The preprocessing steps included removing outliers, handling missing values through imputation, and calculating WQI using weighted hydrochemical parameters. Five machine learning models were evaluated for performance: logistic regression, decision tree, support vector machine (SVM), naïve Bayes, and a feedforward neural network also known as a multilayer perceptron (MLP). The neural network model achieved the highest classification accuracy of 93%, followed by logistic regression at 89%. To further enhance performance, a custom stacking ensemble model named Neuro‐StackNet was developed. This model integrates the neural network and logistic regression as base learners while using random forest as the meta‐learner. Neuro‐StackNet achieved an accuracy of 95% and effectively classified the water samples into three quality categories: Excellent, Good, and Poor. This study offers a scalable, data‐driven framework for urban groundwater quality assessment. It can support policymakers in identifying contamination hotspots and planning targeted remediation strategies.