Improving the bankruptcy prediction by combining some classification models

Bankruptcy prediction problem has been a concerned topic due to its negative effects to areas of life namely economy, society and policy. However, once this problem being solved, it will help to repulse the above effects and helps individuals and organizations to make wise decision. The purpose of predicting bankruptcy is to develop a predictive model that allows to foreseen likelihood of financial distress of company. In this paper, we consider "Polish companies' bankruptcy dataset "that includes financial ratios and the bankruptcy situation of 43405 Polish companies. In literature, the best result on the dataset reaches a AUC of 95.9%. The difficulties of the dataset come from the huge number of missing values and the imbalance between majority and minority class. In our approach, we used some techniques including random forests regression and Synthetic Minority Oversampling Technique(SMOTG) method to treat he dataset. Some classifiers and a proposed combination of three best classifiers were then applied and evaluated by the cross-validation method. The results show that the proposed method has a good performance with a AUC of 99.78%, a F1 score of 95.12%.