Landslide spatial prediction based on cascade forest and stacking ensemble learning algorithm

Landslides are a major threat to the safety of human life and property. The purpose of landslide spatial prediction is to establish the relationship between the location of landslides and each landslide evaluation factor, and to spatially identify high landslide risk areas using data mining and geographic information science. In this paper, a landslide spatial prediction model is put forward based on cascade forest (CF) and Stacking ensemble learning algorithm. Firstly, the landslide spatial prediction scheme is designed. Then, the improved CF is established by combining random forest (RF) and extreme gradient boosting (XGBoost). The Stacking ensemble learning algorithm is introduced to establish CF-Stacking model combined with the improved CF. Finally, experiments are conducted using geospatial data of the actual study area. 12 landslide disaster-inducing factors are extracted from the study area, and the CF-Stacking model is applied to the spatial prediction of landslides. The result shows that CF-Stacking outperforms comparative models in terms of the area under curve and brier score, demonstrating its effectiveness in predicting landslide spatial patterns. The CF-Stacking model is used to generate a landslide susceptibility map for Fengjie, which provides valuable guidance for geological hazard early warning.