Artificial neural network and sensitivity analysis in the landslide susceptibility mapping of Idukki district, India

Abstract

Idukki district faced adverse mishappenings during the 2018 Kerala landslides due to incessant torrential rainfall. This study emphasizes developing an efficient and accurate ANN model to integrate the data, process and generate landslide susceptibility maps. Fifteen conditioning factors that influence landslides' occurrence opted in the study constitutes 49 input neurons to the ANN model (L49). Seven inputs with high robustness were identified using the sensitivity analysis approach and were adopted to generate a new ANN model (L7). Both ANN models were processed to obtain an optimal output with lesser cross-entropy error. The landslide susceptibility maps derived from these ANN models show similar trends with the region's observed landslide locations. The ANN models were validated using ROC, and it provided a very good fit with AUC values of 0.91 and 0.83 as prediction rate for ANN models L49 and L7, respectively.