Development of decided tree model and alternating decision tree models for spatial prediction of landslides at Muong Nhe district, Dien Bien Province, Vietnam

Abstract

Landslide spatial prediction map is a useful tool for effective management of landslides of an area. In this study, we adopted machine approach based on alternating decision tree (ADT) and decision tree (DT) to predict landslides spatially at Muong district, Dien Bien Province, Vietnam. In this study, a total of 159 landslide past landslides were identified and 12 conditioning factors including: slope, aspect, curvature, elevation, distance to faults, NDVI, flow accumulation, TWI, SPI, geological, distance to rivers, distance to roads were used to build spatial database for landslide spatial
modeling. Validation and comparison of the models was done using statistical indicators including ROC / AUC curves. The results show that performance of both ADT and DT is good for spatial prediction of landslides but ADT model (AUCtrain = 0.928, AUCtest = 0.887) is better than DT model (AUCtrain = 0.915, AUCtest = 0.800). Map of landslide spatial prediction at Muong Nhe District can be used in land use planning for better landslide hazard management.