This study aims to develop three novel GIS-based models combining
Genetic Algorithm (GA), Biogeography-Based Optimization (BBO) and
Simulated Annealing (SA) with Support Vector Regression (SVR) for
groundwater potential (GP) mapping in the governorate of Tafillah,
Jordan. Twelve topographical, hydrological and geological factors
were considered. The mapping process was done with and without
feature selection (FS) conducted by integration of SVR model with
GA, BBO and SA algorithms. The accuracy of these models was evaluated using the area under receiver operating characteristic (AUROC)
curve. Comparisons among the models uncovered that the SVR-RBFGA and SVR-RBF-BBO models performed better than the SVR-RBF-SA.
The AUROC for two mentioned models were 0.964 and 0.996 in training and testing runs, respectively, while this metric was 0.953 and
0.986 for SVR-RBF-SA model in training and testing runs, respectively.
The results showed that after FS, the models are more accurate in test
data than train data.
