Enhancing Rural Water Conservation through Deep Learning Based Multi-Class Water Quality Classification

Ensuring sustainable water use and safeguarding public health require continuous monitoring of water quality, particularly in rural and underserved regions. However, real-world water quality classification is frequently hindered by severe data imbalance, where rare but hazardous contamination events are underrepresented in training data, leading to poor detection accuracy and increased environmental risk. Using sensor data from the rural drinking water system in Heingang, Manipur, this study classifies water quality into five categories based on the Indian Water Quality Index (WQI). To effectively address the multi-class imbalance in real-time water quality classification, a deep ensemble framework, DNN-CW-BBAG (Deep Neural Network-Class Weights-Balanced Bagging), is proposed that combines class weighting with balanced bagging, offering both scalability and robustness. While previous studies have suggested deep learning or ensemble methods independently for water quality analysis, this study integrates both to develop a practically viable solution tailored for real-world imbalance scenarios. The model’s generalization ability was validated using a separate dataset from Lalpura, Rajasthan, where it maintained high predictive performance, reinforcing its adaptability across geographically diverse regions. This work supports informed decision-making for risk management, community-level conservation, and sustainable water usage, directly contributing to Clean Water and Sanitation. Although effective, the model’s broader applicability is limited by data availability. Future work should incorporate more diverse datasets and lightweight deployment strategies to enhance transparency and scalability in field-based water quality surveillance.