A Hybrid Machine Learning Framework for Water Quality Index Prediction Using Feature-Based Neural Network Initialization

This project presents an advanced machine learning framework for water potability prediction aimed at supporting intelligent water quality monitoring and public health management. The framework utilizes physicochemical water quality indicators, including pH, Hardness, Solids, Chloramines, Sulfate, Conductivity, Organic Carbon, Trihalomethanes, and Turbidity, to determine whether water is safe for human consumption. A comprehensive preprocessing pipeline was implemented, involving missing value handling, feature scaling, data balancing, and train-test splitting to ensure robust model performance. The proposed framework introduces a novel Adaptive Attention-Guided Hybrid Ensemble (AAGHE) model that combines deep latent feature extraction through an Autoencoder with multiple ensemble learners, including CatBoost, Extra Trees, and HistGradient Boosting classifiers. An attention-guided neural meta-learner is employed to intelligently fuse predictions from the base models, enhancing classification accuracy and generalization capability. For comparative evaluation, a Dual-Transformer Cross-Fusion Network (DTCF-Net) was also developed to capture complex feature interactions using transformer-based architectures. Experimental results demonstrate the superior performance of the proposed AAGHE framework, achieving an accuracy of 99.33%, with precision, recall, and F1-scores exceeding 99%, significantly outperforming the DTCF-Net model, which achieved an accuracy of 68.81%. The trained models were saved for deployment and real-time inference applications. By effectively modeling complex nonlinear relationships among water quality parameters, the proposed framework provides a reliable and scalable solution for automated water safety assessment. This approach can assist environmental agencies, water treatment facilities, and policymakers in ensuring safe drinking water, enabling proactive decision-making, efficient resource management, and sustainable water quality monitoring.

Keywords: Water Potability Prediction, Water Quality Assessment, Adaptive Attention-Guided Hybrid Ensemble (AAGHE), Autoencoder, CatBoost, Extra Trees, HistGradient Boosting, Deep Learning, Transformer Networks, DTCF-Net, Machine Learning, Environmental Monitoring, Public Health, Predictive Analytics.