Modeling of Ground Water Quality Index of Haryana: A Comparison of Deep Neural Network and Support Vector Machine

Deep Neural Network (DNN) and Support Vector Machine (SVM) machine algorithms have been applied for evaluating the water quality index (WQI) over the Haryana state (India). The data has been collected from the National Ground Water Monitoring Stations distributed in twenty districts of Haryana during the period of May 2015. The model was tested using 62 groundwater samples, leaving 247 randomly chosen samples from a total of 309 groundwater samples. Fourteen input variables consist of pH, TH, EC, Na+, Mg2+, Ca2+, K+, Cl−, SO42−, HCO3−, NO32−, CO32−, F−, SiO2 and the computed value of WQI was taken as output. These statistical coefficient variables such as Correlation coefficient (CC), Root mean square error (RMSE) and Mean absolute error (MAE) were calculated to compare the performance of two modeling approaches. The result based on various indicators suggests that DNN works better than the SVM in predicting water quality. Keeping in view of improved performance by the DNN model, this approach can be proposed as an alternate to other state of art modeling approach (such as SVM) in predicting groundwater quality successfully.