Predictive modeling of water quality index (WQI) classes in Indian rivers: Insights from the application of multiple Machine Learning (ML) models on a decennial dataset.

Despite high pollution levels in Indian rivers, a comprehensive study on the water quality index (WQI) remains elusive. WQI values were computed, and their classes were determined using six water quality parameters from an available decennial dataset (n = 3595) on Indian Rivers. This study aims to assess the spatial distribution of WQI values and their classes across Indian River systems while exploring the application of machine learning (ML) based models in predicting WQI classes using a reduced number of input parameters. Modeling experiments were designed on five models- Decision Tree (DT), Random Forest (RF), Gradient Boosted Trees (GBT), Artificial Neural Network (ANN), and Support Vector Machine (SVM) for predicting WQI classes. Each model was trained with input parameters and WQI classes with 2990 datasets. Testing of WQI classes by each model was made on 605 datasets under different framework sets. Models' performance metrics were evaluated by accuracy, weighted mean recall and precision, and F-score. Our study demonstrates that the two largest systems, Ganga and Brahmaputra, lie on the extremes of the WQI (mean) spectrum, reflecting the impact of contrasting population density, industrial activities, change in land-use-land-cover pattern, and agricultural use on the riverine WQI. Our modeling experiments underscore that with only three input parameters, GBT can predict WQI classes with > 80% of performance metrics. With only two input parameters, GBT, RF, and ANN, all can provide reliable estimates. Our study highlights that ML models can serve as decision-supporting tools for water resource policymakers and managers in making effective pollution control and water resource management decisions. [ABSTRACT FROM AUTHOR]