Landslide susceptibility assessment based on the coupling of information value and machine learning models

Guizhou Province’s geological environment is particularly fragile, and the landslide-prone environment is complex, Landslide geological disasters are widely distributed. Taking Tongzi County as the study area, this research constructs an evaluation system covering five types of evaluation factors: topography and geomorphology, basic geological environment, meteorological and hydrological conditions, human activities, and vegetation coverage. The Support Vector Machine (SVM) and the eXtreme Gradient Boosting (XGBoost) models are each combined with the Information Value model (IV). The optimal grid unit is introduced to obtain the optimal parameter combinations of the single models and their coupled models for model training. Ultimately, SVM, XGBoost, IV-SVM, and IV-XGBoost—were established for the purpose of assessing landslide susceptibility in Tongzi County, with validation carried out via the Receiver Operating Characteristic (ROC) curves. For these four models, the Areas Under the Curve (AUC) corresponding to their ROC prediction curves are 0.75, 0.82, 0.89, and 0.94 in sequence. The findings indicate that the coupled models achieve higher accuracy than the standalone models. IV-XGBoost model exhibits superior performance and higher accuracy, with landslides predominantly concentrated in the southern and northwestern mountainous road sections and water-adjacent regions of the Tongzi County—areas marked by a complex geological setting and frequent human activities. This study seeks to offer a robust scientific foundation for the forecasting and prevention of landslide hazards in Tongzi County, as well as a practical and efficient approach for landslide susceptibility assessment research.