Improvement of landslide spatial modeling using machine learning methods and two Harris hawks and bat algorithms

Abstract Landslide is a natural phenomenon that can turn into a natural disaster. The main goal of this research was to spatial prediction of a high-risk region located in the Zagros mountains, Iran, using hybrid machine learning and metaheuristic algorithms, namely the adaptive neuro-fuzzy inference system (ANFIS), support vector regression (SVR), the Harris hawks optimisation (HHO), and the bat algorithm (BA). The landslide occurrences were first divided into training and testing datasets with a 70/30 ratio. Fourteen landslide-related factors were considered, and the stepwise weight assessment ratio analysis (SWARA) were employed to determine the correlation between landslides and factors. After that, the hybrid models of ANFIS-HHO, ANFIS-BA, SVR-HHO and SVR-BA were applied to generate landslide susceptibility maps (LSMs). Finally, in order to validation and comparison of the applied models, two indexes, namely mean square error (MSE) and area under the ROC curve (AUROC), were used. According to the validation results, the AUROC values for the ANFIS-HHO, ANFIS-BA, SVR-HHO and SVR-BA were 0.849, 0.82, 0.895, and 0.865, respectively. The SVR-HHO showed the highest accuracy, with AUROC of 0.895 and lowest MSE of 0.147, and ANFIS-BA showed the least accuracy with an AUROC value of 0.82 and MSE value of 0.218. Based on the results, although four hybrid models with more than 80% accuracy can generate very good results, the SVR is superior to the ANFIS model, whereas the HHO algorithm outperformed the bat algorithm. The map generated in this study can be employed by land use planners in more efficient management.