Introducing a Hybrid Model Based on Certainty Factor Statistical and Bagging Methods for Discovering Groundwater

Due to climate change and growth of urban communities, the need for groundwater and exploration of these resources are increasing. Therefore, the purpose of this study was to provide a groundwater potential mapping using the geographic information system (GIS) in a region located in Booshehr plain using an ensemble of certainty factor (CF) method with Bagging data mining method. For this purpose, in the first step, 339 well locations were identified in the study area, of which 238 wells (70%) were randomly selected as training points and 101 wells (30%) were selected as validation points. In the next step, 15 factors affecting groundwater such as altitude, slope angle, slope direction, slope length, plan curvature, profile curvature, topographic wetness index, distance from fault, fault density, distance from river, drainage density, rainfall, lithology, Soil and land cover were prepared in ArcGIS 10.3 and Saga GIS software. The spatial relationship between the effective parameters and the location of the wells was determined using a CF model. These weights were used to implement the Bagging model. In order to validate the accuracy of the ensemble model, the RMSE and MAE indices were used. Also, in order to validate the accuracy of the maps, ROC and AUC were used. The results of this study showed that the values of RMSE and MAE indices for training and validation are equal to 0.247, 0.162, 0.256 and 0.169 respectively. The evaluation results of the ROC curve indicated that the AUC was 86.2 and 94.8%, respectively, for CF models and the ensemble of CF model with the Bagging model.