Groundwater quality classification using machine learning and cluster analysis ;case of Upper and Mid

Abstract

Monitoring water quality is essential for resource protection and management. This study examines the application of machine learning methods, particularly Support Vector Machine (SVM) and Extreme Gradient Boosting (XGBoost), alongside K-means clustering, to assess groundwater quality in the Upper and Middle Cheliff plains based on WHO (2017) standards. The methodology included data pre-processing and standardization, followed by classifying samples into quality categories for Water Quality Index (WQI) computation. SVM and XGBoost models underwent training and evaluation through stratified cross-validation, utilizing performance metrics such as accuracy, precision, recall, and F1-score, along with Kmeans for clustering. Results showed that XGBoost outperformed SVM with 82.98% validation accuracy during high water periods, attributed to its capability in modelling nonlinear relationships and variable importance. Nitrates, chlorides, and EC were identified as pivotal parameters influencing classification. In contrast, during low water periods, SVM outperformed XGBoost with an accuracy of 82.79% compared to 66.73%. The proposed machine learning strategy offers a scalable framework for similar arid and semi-arid regions facing groundwater challenges.