Machine learning aided multiclass classification, regression, and cluster analysis of groundwater quality variables congregated from the YSR district

Abstract

In this study, machine learning classifiers are integrated with the geostatistical analyses. The data extracted from the surface maps derived from ordinary kriging were passed onto ML algorithms, resulting in prediction accuracies of 95% (Gradient Boosting Classifier) for classification and 91% (Random Forest Regressor) for Regression. Kmeans clustering model provided better results in clustering analysis based on Silhouette, Calinski-Harabasz, and Davies-Bouldin metrics. However, there was certain overfitting in the prediction, probably due to limited data available for analysis. In addition, the interpolation methods might have affected the model performance by producing overfitting and underfitting results. It is to report that the Gradient Boosting classifier in classification mode yielded relatively high prediction accuracies in predicting groundwater quality when three classes were used. The Random Forest Regressor in regression mode returned better results in predicting groundwater quality features when multiple classes were used in this study. This work reports that machine learning algorithms can predict groundwater quality with minimal expense and expertise.