A novel predictive analysis approach for forecasting and classifying surface water data using AWQI standards and machine learning-based rule induction

Preserving surface water is requisite as it is a critical natural resource. As populations grow, many rising nations, like India, have substantial issues in controlling surface water contamination. Inadequate operation and maintenance of Sewage/ Effluent Treatment Plants (STPs/ETPs), as well as the absence of dilution and other non-point source factors, contribute to water effluence issues across the nation. Neglected and partially treated wastewater from municipalities and industrial sources flow into waterways, further exacerbating the problem. Therefore, there is a compelling need to investigate the presence of harmful substances in the water and to identify regions with higher concentrations of these pollutants. When given the data on the chemical components of the water, Water Quality Index (WQI) models and Machine Learning (ML)-based methods have shown to be a superior substitute for analyzing and predicting the quality of the water. However, these models are time consuming due to the increased parameter count depending on sub-index calculation, prediction time and tuning for model evaluation. So, a novel predictive analysis methodology for determining the rules based on the Assam Water Quality Index (AWQI) norms is proposed to address this problem with least number of attributes. Dissolved Oxygen (DO), Biological Oxygen Demand (BOD), Fecal Coliform (FC), and Total Coliform (TC) are selected in the proposed model to derive rules. The Assam Water Quality Classification (AWQC) scheme is used to classify surface water quality after the rules have been created. In addition, performance of the proposed approach is compared with the existing models Random Forest (RF), Extreme Gradient Boosting (XGBoost) and Decision Tree (DT) in terms of effective metrics. The novel predictive approach performs optimally, with an accuracy of 0.99%, precision of 0.98%, recall of 100%, f1-score of 0.99%, AUC of 0.99%, and classification error of 0.008%. This proposed model will improve the capabilities and effectiveness of predictive systems, allowing it to resolve a broader range of difficulties.