Landslide susceptibility assessment using information quantity and machine learning integrated models: a case study of Sichuan province, southwestern China

Landslides represent a significant natural hazard in Sichuan Province, causing considerable property damage and loss of life. Utilizing machine learning (ML) for landslide susceptibility mapping (LSM) is an effective strategy for mitigating landslide risk. Nonetheless, individual ML models face certain limitations in interpreting landslide data, assigning weights to evaluation factors, and achieving high predictive accuracy. These limitations hinder the precision of landslide susceptibility classifications worldwide. To overcome these issues, this research proposes a hybrid ML model that combines information value (IVM) with ML, referred to as IVM-ML, to improve the precision of landslide susceptibility prediction (LSP). The dataset includes 7,840 historical landslide events and incorporates 14 landslide evaluation factors (LEFs), which were analyzed for their suitability through correlation and multicollinearity assessments. Low-susceptibility zones identified by the IVM model were used to select non-landslide samples, and six models—SVM, RF, LR, IVM-SVM, IVM-RF, and IVM-LR—were employed for LSP. The effectiveness of these models was evaluated based on Sensitivity (Se), Accuracy (Ac), Specificity (Sp), Precision(Pr), F1 Score(F1), Frequency ratio of landslide sites(Fr) and the area under the ROC curve (AUC). The findings reveal that the southeastern and eastern regions of the research region, which encompass approximately 30% of the entire area, exhibit a higher risk of landslides, while the western and northwestern areas, comprising approximately 45%, have a lower risk. The IVM-ML models, particularly IVM-RF, achieved notably higher predictive accuracy, with AUC values of 0.997, 0.996, and 0.998 for IVM-SVM, IVM-RF, and IVM-LR, respectively, outperforming the standard ML models. These results highlight the IVM-ML model’s potential for improving LSP accuracy, particularly in high-risk regions, contributing significantly to landslide hazard mitigation in Sichuan and globally.