Predicting landslide and debris flow susceptibility using Logitboost alternating decision trees and ensemble techniques

Landslides are a global hazard that requires smart tools to identify the most vulnerable areas and to implement effective prevention and recovery plans. This study developed three ensemble models to assess the spatial susceptibility of landslides and debris flow in the Nam Pam commune of the Son La province, Vietnam. We applied the LogitBoost alternating decision trees (LADT) method as the base classifier and combined it with Bagging (B), Dagging (D), and MultiBoost (MBAB) ensemble techniques as ensemble techniques. We collected the locations of past landslides and debris flows from extensive field surveys and related them to sixteen variables that thought to influence landslide and debris flow occurrence to examine the spatial distribution of landslide and debris flow susceptibility in the study area. The models were evaluated based on the area under the receiver operating characteristic curve (AUC) and other evaluation criteria, such as positive predictive value (PPV), negative predictive value (NPV), sensitivity (SST), specificity (SPF), accuracy (ACC), Kappa, and root mean square error (RMSE). The results showed that the B-LADT model was the best model, with AUC = 0.9, PPV = 86%, NPV = 82%, SST = 83%, SPF = 86%, ACC = 85%, RMSE = 0.36, and Kappa = 0.69. According to this model, about 17% of the study area had high and very high landslide and debris flow susceptibility levels. These regions were mainly associated with the variations in weathering crust, elevation, fault density, and lithology of the study area. The study demonstrates the effectiveness of ensemble learning techniques in creating reliable prediction models, which can help save lives and reduce infrastructure damage in landslide- or debris flow-affected regions worldwide.