Machine Learning Methods of Kernel Logistic Regression and Classification and Regression Trees for Landslide Susceptibility Assessment at Part of Himalayan Area, India

Abstract

Objectives: To evaluate performance of machine learning methods for assessment of landslide susceptibility at Himalayan area, India. Methods/Statistical analysis: Machine learning methods namely Kernel Logistic Regression (KLR) and Classification and Regression Trees (CART) were applied and compared in this study. Landslide affecting parameters and 930 historical landslides were used for generating datasets. Receiver Operating Characteristic (ROC) curve and Statistical analysis methods were used for validation and comparison. Findings: Result analysis shows that both the KLR and CART models perform well for landslide susceptibility assessment but the KLR model (AUC = 0.894) outperforms the CART model (AUC = 0.842). Thus, both these methods can be considered as promising machine learning techniques for landslide susceptibility assessment; however, the KLR is better than the CART. Application/Improvements: Results of this study would be useful for susceptibility assessment and landslide hazard management in landslide prone areas.