Meta optimization of an adaptive neuro-fuzzy inference system with grey wolf optimizer and biogeography-based optimization algorithms for spatial prediction of landslide susceptibility

Abstract

Abstract Estimation of landslide susceptibility is still an ongoing requirement for land use management plans. Here, we proposed two novel intelligence hybrid models that rely on an adaptive neuro-fuzzy inference system (ANFIS) and two metaheuristic optimization algorithms, i.e., grey wolf optimizer (GWO) and biogeography-based optimization (BBO), for obtaining a reliable estimate of landslide susceptibility. Sixteen causative factors and 391 historical landslide events from a landslide-prone area of the State of Uttarakhand, northern India, were used to generate a geospatial database. The ANFIS model was employed to develop an initial landslide susceptibility model that was then optimized using the GWO and BBO algorithms. This resulted in two novel models, i.e., ANFIS-BBO and ANFIS-GWO, that benefited from an intelligent approach to automatically and properly adjust the best parameters of the base ANFIS model for the prediction of landslide susceptibilities. The robustness of the models was verified through a large number of runs using different splits of training and validation datasets. Although few differences observed between the predictive capability of the models (AUCANFIS-BBO = 0.95; RMSEANFIS-BBO = 0.316 vs. ACUANFIS-GWO = 0.94; RMSEANFIS-GWO = 0.322), the Wilcoxon signed-rank test indicated a significant difference between the model performances in both training and validation datasets. Overall, our proposed models demonstrated an improved prediction of landslides compared to those achieved in previous studies with other methods. Therefore, these novel models can be recommended for modeling landslide susceptibility, and the modelers can easily tailor their use based on their individual circumstances.