To achieve accurate prediction of the flight punctuality rate, a flight regularity prediction index system was constructed based on data statistics of flight delay reasons, includeing departure airport, destination airport, flow control information, and route characteristics. It proposes a SMOTE algorithm-based XGBoost classification prediction model (SM-XGBoost model) and a SMOTE algorithm-based LightGBM classification prediction model (SM-LightGBM model). Based on the actual data of major airports in East China, the validity and progressiveness of the proposed model are verified. The results showed that the SM-XGBoost model and SM-LightGBM model were significantly better than the decision tree and random forest models in terms of prediction accuracy and error. In terms of stability of training set and test set, SM-LightGBM model is superior to the SM-XGBoost model, with a maximum prediction accuracy of 88.2% for test set. This method provides a new analytical approach for predicting events in similar complex systems.