Decision tree-based data mining and rule induction for identifying hydrogeological parameters that influence groundwater pollution sensitivity

This study aims to develop a new field-based approach that can estimate patterns of groundwater pollution sensitivity using data mining algorithms. Hydrogeological and pollution sensitivity data were collected from the Woosan Industrial Complex, Korea, which is a site contaminated by trichloroethylene (TCE). The proposed data mining algorithm procedure uses seven hydrogeological properties as input variables: depth to water, net recharge, aquifer media, soil media, topography, vadose zone media, and hydraulic conductivity. The observed TCE sensitivity was used as the target data. Initially, four data mining algorithms artificial neural network (ANN), decision tree (DT), case-based reasoning (CBR), and multinomial logistic regression (MLR) were tested. We found that the DT-based data mining and rule induction method shows better prediction accuracy and consistency than the other methods. We also used the ordinal pairwise partitioning (OPP) algorithm to improve the accuracy and consistency of the DT model. A classification and regression tree (CART) analysis of the OPP-DT model indicated that the net recharge (R), soil media (S), and aquifer media (A) were the major hydrogeological factors that influence groundwater sensitivity to TCE at the site. The results of this study demonstrate that the proposed model can provide more accurate and consistent estimates of groundwater vulnerability to TCE compared to the existing models.