Abstract
Groundwater scarcity is one of the most concerning issues in arid and semi-arid regions. In this study, we develop and validate a novel artificial intelligence that is a coupling of five ensemble benchmark algorithms e.g., artificial neural network (ANN), reduced-error pruning trees (REPTree), radial basis function (RBF), M5P and random forest (RF) with particle swarm optimization (PSO) for delineating GWP zones. Further, nine parameters used for the GWP modelling and to test and train the proposed PSO-based models. Additionally, this study proposes a receiver operating characteristic (ROC) based sensitivity analysis for GWP modelling. Multicollinearity test, information gain ratio, and correlation attribute evaluation methods used to choose important parameters for the proposed GWP model. The result shows that drainage density, elevation, and land use/land cover have a higher influence on the GWP using correlation attribute evaluation methods. Results showed that the hybrid PSO-RF model performed better than other proposed hybrid models.