Recursive Feature Elimination for Machine Learning-based Landslide Prediction Models

Abstract

This paper proposes a landslide prediction model which uses the recursive feature elimination method, which is one of the key feature selection methods in machine learning that is not tested yet for landslide prediction related applications. The model is tested with the landslide inventories of two landslide-prone areas. The results show that the proposed model achieves an average accuracy of 91.15% and a sensitivity of 83.4% in predicting the possibility for a landslide. The findings of this research paper imply that recursive feature elimination can also be effectively used in landslide predictions since it achieves high accuracy.