Forest fires pose a significant threat to human lives and property. Accurate prediction of forest fire risk is crucial for disaster mitigation and prevention. Influenced by factors such as terrain, meteorology, vegetation cover, and human activities, the causes of forest fires exhibit regional differences. This study uses historical forest fire events in Muli County, Sichuan Province as the response variable, with terrain, meteorological data, vegetation cover, and human activity data as explanatory variables. Leveraging CatBoost's strengths in handling high-dimensional sparse data and classification problems, a high-precision forest fire prediction model was constructed based on CatBoost. The experimental results indicate that, compared to random forest (RF), extreme gradient boosting(XGBoost), and gradient boosting decision trees(GBDT) models, the CatBoost model achieves higher modeling accuracy and significantly improves forest fire prediction accuracy, with a prediction accuracy rate of 91.36% and an area under curve(AUC) value of 0.970. Predictions made using this model can provide valuable references for the early prevention of forest fires in Muli County.