The applications on the assessment and analysis of the susceptibility of geohazards using machine learning is one of the hotspots in current researches. To evaluate the geohazards susceptibility reasonably and effectively, the model of eXtreme Gradient Boosting (XGBoost) is used by coupling with geographic information system (GIS) technology to carry out multi-classification evaluation of landslide geohazards susceptibility, which is based on the geological disaster survey data of Feiyun River basin in Wenzhou, Zhejiang Province. The input data include slope, slope aspect, slope shape, landcover type, Topographic Wetness Index (TWI), extreme hourly rainfall, internal friction angle, cohesion, bulk density and regolith thickness 10 landslide hazard factors. The model results were evaluated by multi-classification confusion matrix and compared with Support Vector Machine (SVM) model. The results show that the recall rate and accuracy rate of the trained XGBoost algorithm model for the identification of extremely high risk areas in the test set have reached 97.92% and 98.06%, respectively, and the F1 value has reached 97.99%, which are both better than SVM model. Therefore, the proposed and trained model can provide model support for the susceptibility evaluation of landslide geological disasters in the studied area.