<span style="mso-fareast-font-family: 'DengXian Light'; mso-fareast-theme-font: major-fareast;">Comparative Study of Regional Landslide Susceptibility Evaluation by the Information Quantity Method and Machine Learning Coupled Modeling with a Case of Maerkang City

This study focuses on the landslide susceptibility assessment in Maerkang City, Sichuan Province. Twelve evaluation factors, including terrain relief and slope, were selected for analysis. The study employs the Information Value-Analytic Hierarchy Process (IV-AHP), Random Forest (RF), Extreme Gradient Boosting (XGBoost), as well as hybrid models IV-RF and IV-XGBoost to evaluate landslide susceptibility and explore the differences between traditional methods, machine learning models, and hybrid models. The findings indicate that traditional statistical analysis models exhibit relatively low prediction accuracy. While machine learning models (RF and XGBoost) improve accuracy, they suffer from overfitting and poor interpretability. To address these issues, this study adopts hybrid models that integrate the strengths of both traditional statistical methods and machine learning approaches, demonstrating superior accuracy in landslide susceptibility assessment. Based on field survey data and multiple model predictions, it was observed that in high-altitude areas of western Sichuan Province, landslides tend to exhibit a certain degree of concealment, making them difficult to detect during data collection. This leads to discrepancies between sample data and actual conditions, thereby affecting the accuracy of prediction results. The findings provide a reliable scientific basis for landslide disaster prevention and management, while also highlighting future research directions, including the application of hybrid models, the enrichment of sample data, and the analysis of the applicability of different assessment methods.