Abstract: To address the challenge of determining the weights of evaluation factors in geological disaster susceptibility assessment due to insufficient sample data, this study proposed a geological disaster susceptibility prediction model based on transfer learning strategy, taking Chengguan Town of Zhenping County in Shaanxi Province as a case study. The TrAdaBoost algorithm was employed, taking disaster data from Zhenping County as the source domain with data from the target region (Chengguan Town). CatBoost was used as the base learner to establish the TrAdaBoost-CatBoost geological disaster susceptibility model.Results show TrAdaBoost-CatBoost model significantly improved AUC value from 0.94 to 0.96 compared with CatBoost model using only Chengguan Town data. Additionally, TrAdaBoost-CatBoost model outperformed traditional support vector machine (SVM) model and random forest (RF) model in ROC curve performance with the AUC of TrAdaboost-CatBoost model of 0.96, the AUC of SVM model of 0.92, and the AUC of RF model of 0.93. The migration framework was universal, the AUC of TrAdaBoost-SVM model was 0.94, and the AUC of TrAdaBoost-RF model was 0.95. Their performance was both improved, but TrAdaBoost-CatBoost model still remained optimal. The findings could provide valuable insights for geological disaster susceptibility assessment in scenarios with limited disaster samples.