Abstract: In view of the low accuracy and accuracy of the existing fault prediction models of rolling bearing, a rolling bearing fault prediction model based on ARIMA time series prediction and XGBoost classification algorithm is proposed in this paper. Firstly, LMD combined with FPA is used to solve the problem of blind source underdetermination. Secondly, KPCA is used to select sensitive features as the input of the prediction model to improve the classification accuracy of bearing faults. Thirdly, Arima autoregressive model is used to predict the short-term change of bearing vibration signal in the future, and the prediction results are input into XGBoost model for fault classification and prediction, so as to realize rolling bearing fault identification and improve the prediction accuracy. Finally, an example is verified by the bearing data set used by Case Western Reserve University in the United States. The experimental results show that this method can more accurately predict the short-term vibration signal change of the bearing and diagnose the possible faults. It is proved that this method can effectively extract the features under the condition of rolling bearing signal noise. It is feasible and reliable in fault identification and fault early warning.