Geostatistical analyses empowered with gradient boosting and extra trees classifier algorithms in the prediction of groundwater quality and geology-lithology attributes over YSR district, India

Abstract

Machine learning classifiers are integrated with the geostatistical analyses through interpolation techniques to predict groundwater quality and geology-lithology. Ordinary kriging is used to obtain the optimal interpolation model using RMSSE values. The data extracted from the surface maps were passed onto ML algorithms, resulting in prediction accuracies of 99% for groundwater quality and 96% in predicting the geology-lithology features. There was certain overfitting in the prediction and it was probably due to several classes of geology-lithology variables and limited data available for analysis. The interpolation methods might affect the model performance by producing overfitting and underfitting results. It is to report that the gradient boosting classifier yielded relatively high prediction accuracies in predicting groundwater quality when two classes were used. The extra trees classifier returned better results in predicting geology-lithology features when multiple classes were used in this study.