Financial data is an important tool for the construction of models of financial distress. As the amount of data increases, machine learning-based models perform better than traditional financial distress models. However, in the real world, the classification performance of machine learning will be reduced due to the class imbalance of the data. This paper applies hybrid ensemble methods to address class-imbalance problems of binary financial datasets in machine learning. It integrates a data augmentation module employing oversampling techniques and LightGBM algorithms. A comparative analysis of the model performance, as measured by the seven evaluation metrics. From the average results of the 15 datasets, Our method balances the classification accuracy of minority classes with overall performance. It alleviates the issue of the traditional method of improving recall rates sacrificing classification accuracy. The hybrid model integrating the data augmentation module and the LightGBM algorithm achieved an accuracy of 98.3%. Precision (76%), Recall (61.98%), F1-score (0.6664), Kappa (0.6582), MCC (0.6698), and AUC (0.8071) were the best among the seven models. The computational efficiency of the proposed method is second only to the decision tree model with simple structure while ensuring high performance.