Abstract: The causes of mine karst surface collapse are complex and diverse. To accurately predict karst collapse in mines, combined with the karst development mechanism, a BP neural network prediction model based on the LOF and SMOTE algorithms is proposed. In this model, the abnormal data due to unnatural reasons were first removed by LOF algotithm. The removed data were then oversampled by SMOTE algorithm, thereby synthesizing new data to increase the number of samples. Finally, the BP neural network model was used to predict the mine karst collapse. The results show that the preprocessed prediction model of the actual engineering data has higher prediction accuracy than some small sample prediction models, which provides a reference for its application in other projects.