Evaluation of groundwater quality for drinking purposes based on machine learning algorithms and GIS
Hemant Raheja, Arun Goel, Mahesh Pal
Abstract
Groundwater is one of the most valuable sources of water for drinking use. This study uses 94 groundwater samples collected from the tube wells during pre- and post-monsoon period of 2022 from different locations in Rohtak district, Haryana (India). Fourteen hydrochemical parameters for each sample were determined and compared with the standard values prescribed by World Health Organization (WHO) and Bureau of Indian Standards (BIS) 10,500:2015 for drinking purpose. Drinking Water Quality Index (DWQI) calculated using different hydrochemical parameters was found to vary from 95.02 to 448.92 and 93.91–497.72 during pre- and post-monsoon season, respectively. Spatial distribution maps of different hydrochemical parameters and DWQI indicate that the not-permissible water quality values for drinking were found in the western region of the study area during both seasons. Two machine learning algorithms, including Gaussian Process Regression (GPR) and Support-Vector Regression (SVR) algorithms with four kernel functions, were used to predict DWQI value using pre-monsoon samples (training) and post-monsoon samples (testing). Results suggest an improved performance by the SVR algorithm using Radial basis kernel (SVR-RBF) compared to other kernel functions with both SVR and GPR approaches. Sensitivity analysis indicates that TDS, NO 3 − , and F − are three important parameters for predicting the DWQI. The outcomes of the proposed algorithms will benefit the government authorities and help recommend alternative drinking water in the affected areas.