Deep Multi Target Classification Framework for Water Quality Monitoring and Pollution Severity Level Assessment

Abstract

Water quality evaluation is essential for protecting ecosystems and safeguarding human health. Conventional monitoring techniques primarily rely on manual sample collection followed by laboratory-based chemical and biological testing. Although these methods provide reliable results, they require considerable time, labor, and resources, making them unsuitable for continuous or real-time monitoring and delaying critical environmental decisions. This study introduces an intelligent and automated framework for water quality analysis and pollution prediction by leveraging advanced Machine Learning (ML) algorithms. The proposed solution is developed as a Flaskbased web application that integrates Classification and Regression Tree (CART) techniques for both pollution category identification and Water Quality Index (WQI) estimation. The implemented models include Linear Logistic Regression (LLR), Decision Tree (DT), Random Forest (RF), Gradient Boosting (GB), and Auto-Interpretable TaoTree (AITT). Among the classification models, AITT delivers the highest prediction accuracy for pollution level identification, while its regression variant achieves the best R² score for WQI estimation, demonstrating strong predictive capability. The application accepts 16 water quality parameters as user input and instantly generates pollution classifications along with WQI predictions. In addition, the platform provides Exploratory Data Analysis (EDA) visualizations, comparative model evaluation, and model retraining functionality, enabling environmental professionals to analyze trends and update predictive models using newly available datasets. By integrating CART-based machine learning workflows with the Flask framework, the system streamlines data preprocessing, model development, prediction, and visualization, minimizing manual intervention while improving accuracy and consistency. Built using scikit-learn, imodels, and pandas, the proposed framework offers a scalable, interpretable, and dependable approach to water quality assessment, supporting effective environmental monitoring and informed decision-making.