Urban Water Assessment using Auto Interpretable Learning for Pollution Classification and Quality Index Modeling

Abstract

Water quality assessment plays a crucial role in ensuring environmental safety and public health. Traditional approaches depend on manual sampling and laboratory-based chemical and biological analyses. These methods are labor-intensive, time-consuming, and often fail to provide real-time insights, limiting their applicability for timely decision-making. This research proposes an intelligent, automated framework for water quality monitoring and pollution prediction using modern Machine Learning (ML) techniques. The system is implemented as a Flask-based web application integrated with Classification and Regression Tree (CART) models for both pollution level classification and Water Quality Index (WQI) prediction. The models utilized include Linear Logistic Regression (LLR), Decision Tree (DT), Random Forest (RF), Gradient Boosting (GB), and Auto-Interpretable TaoTree (AITT). For pollution classification, the AITT model achieves the highest accuracy, whereas for WQI regression, the same model, adapted for regression, produces the best R² score, demonstrating robust predictive performance. Users provide 16 water sample parameters through the interface, enabling instant predictions for pollution levels and WQI values. The system also features Exploratory Data Analysis (EDA) visualizations, model performance comparison, and retraining capabilities, empowering environmental engineers to monitor trends and adapt models to new datasets. By combining CART-based ML pipelines with Flask, the system automates preprocessing, model training, prediction, and visualization, significantly reducing manual effort and enhancing reliability. Implemented using scikit-learn, imodels, and pandas, this framework delivers a scalable, interpretable, and accurate solution for water quality assessment, providing actionable insights for environmental management and decision-making.