A Comprehensive Study on Ensemble-Based Imbalanced Data Classification Methods for Bankruptcy Data

In many real world classification problems the data is imbalanced where the distribution of classes is skewed. When the classification data are not approximately equivalent then the classification dataset is imbalanced. For example one class may be extremely low (minority class) and the other class may be extremely high (majority class). This imbalanced nature of data leads the prediction algorithm to be biased towards majority class. The poor representation of minority class affects the performance of the classification algorithm, which is evident through various assessment metrics. In this paper it is suggested to use ensemble techniques for imbalanced datasets, which focuses on binary class problems. The Ensemble bagging and boosting technique is applied on bankruptcy imbalanced data to improve the performance. Experimental study shows that better performance is achieved when SMOTEBOOST and SMOTEBAGGING are used with decision tree, which is a combination of SMOTE and Ensemble bagging and Ensemble boosting algorithm respectively and it outperforms other ensemble techniques.