Abstract: The outburst flood from a barrier dam poses serious threats to the lives, properties, and safety of downstream populations.Accurate prediction of the breach peak discharge is crucial for disaster response.Therefore, this study utilizes data from 55 historical landslide dam breach cases, including dam height (H_d), reservoir storage capacity (V_w), and dam material type, to construct a Support Vector Machine (SVM) model and a Random Forest (RF) model for predicting breach peak discharge.By comparing the model outputs with measured values and empirical formula estimates, the predictive performances of the machine learning models were quantitatively evaluated using two statistical metrics: the coefficient of determination (R²) and root mean square error (RMSE).The results indicated that both of the two machine learning models developed in this study can accurately predict the breach peak discharge of landslide dams.However, the SVM model (R²=0.900, RMSE=0.465) slightly outperforms the RF model (R²=0.857, RMSE=0.556).However, two machine learning models demonstrate higher prediction accuracy than existing empirical formulas.Specifically, compared to the best-performing empirical formula, the SVM model improves R² by 8.7% and reduces RMSE by 23.6%.The findings of this study can provide valuable references for emergency response to barrier dam breach disasters.