Forecasting corporate bankruptcy in imbalanced datasets using a new hybrid machine learning approach

Bankruptcy prediction is a challenging task. Researchers face the problem of class imbalances, because the number of bankrupt firms is much lower than the number of non-bankrupt firms. Resampling methods, which modify data distributions, are commonly employed to deal with this problem. The authors therefore propose a new, alternate, classifier-level solution that combines the adaptive boosting (AdaBoost) algorithm and support vector machine (SVM) methods: Diverse AdaBoostSVM. A comparison of the performance of Diverse AdaBoostSVM, with resampling methods in imbalanced datasets reveal that at moderate degrees of imbalance and in large training sets Diverse AdaBoostSVM is an effective alternative method of predicting bankruptcy, particularly with regard to mid-term forecast horizons.