Improving Bankruptcy Prediction Using Oversampling and Feature Selection Techniques

摘要

There is a continuous interest in finding better methods to predict bankruptcy because many financial decisions can be made based on the result of such methods. The machine learning techniques are increasingly being developed to improve the prediction of bankruptcy. However, because bankruptcy events are relatively few, machine learning techniques accuracy have been unsatisfactory. In this study, we examine the effectiveness of using oversampling to improve the performance of learning algorithms in imbalanced data. We found that oversampling does improve the precision of the learning algorithms. In addition, the study identified the most important attributes that highly correlate to bankruptcy. Lastly, we tested six major ML algorithms for predicting bankruptcy which are Neural Networks, Decision Trees, Random Forests, Support Vector Machine, K-Nearest Neighbor and Logistic Regression. Random Forest, Decision Tree, and KNN were found to be the best techniques for such problem as they produced higher prediction accuracy.