Virtual Booth

Numerous models have been created to forecast the susceptibility to landslides. The models that are produced have some limitations, particularly concerning the issues of overfitting and overestimation. Thus, using the Bayesian hyperparameter Optimization (BayesOpt) algorithm, the goal of this research is to assess and improve the effectiveness of Extreme Gradient Boosting (XGBoost) in predicting the susceptibility of landslides. One of Malaysia’s most landslide-prone areas, Penang Island, was selected as the subject area for this case study. Firstly, this study considered ten Landslide Factors: Digital Elevation Model (DEM), slope angle, slope length, Normalized Difference Vegetation Index (NDVI), Topographic Wetness Index (TWI), Stream Power Index (SPI), distance from the stream, distance from road, plan, and profile curvature that affect landslides. All spatial databases regarding the landslide conditioning elements were created using Geographic Information Systems (GIS). Next, 886 landslides and non-landslide data were randomly separated into 70% training and 30% testing datasets. The results reveal that BayesOpt-XGBoost performed best, with an AUC of 96.20% for Success Rate (SR) and 94.90% for Prediction Rate (PR). The results demonstrated that BayesOpt enhanced Machine Learning (ML) models’ prediction performance and minimized overfitting. The developed landslide susceptibility map showed reliability since there's no occurrence of overestimation. The LSM is essential for selecting sites, designing structures, and preventing disasters.