Abstract: In order to give full play to the advantages of statistics and machine learning model in the analysis and evaluation of wildfire disaster susceptibility, Guilin, which is rich in forest resources and deeply troubled by wildfire disaster, was taken as the research area, and 16 evaluation factors were selected from the aspects of climate, topography, hydrology and humanities. Based on the information value （IV） model, 4 machine learning （ML） models, including logistic regression （LR）, artificial neural network （ANN）, random forest （RF）, and extreme gradient boosting （XGBoost）, were coupled to evaluate the susceptibility of wildfire hazards in Guilin City. The results indicate that the IV-XGBoost model achieved an AUC of 0.957 and an accuracy of 0.921, demonstrating its superior predictive performance. It can effectively assess the susceptibility of wildfire disasters and provide valuable insights for local wildfire prevention and control.

Key words: wildfire susceptibility assessment, information value model, machine learning model, wildfire disaster, factor importance analysis