Bankruptcy Prediction Model Using an Enhanced Boosting Classifier based on Sequential Backward Selector Technique

Corporate bankruptcy prediction is one of the most crucial issues that impact the economic field, both on the local and global scale. The primary purpose of bankruptcy prediction is to investigate the economic state of any corporation and evaluate its distress level. Several machine learning and deep learning models have been used to predict financial failure. However, there is still no technique that resolves all the problems faced in this field. As such, we propose a machine learning model that constitutes a feature selection phase and a classification phase to predict corporate bankruptcy. This technique combines the sequential backward selector (SBS) with AdaBoost and JRip algorithms. The first phase uses SBS to select the best subset of features for the training. The second phase trains the AdaBoost with the JRip classifier to predict each target class. This model is evaluated using the highly imbalanced Polish bankruptcy dataset. The comparative analysis of our model with other techniques proves the efficiency in predicting corporate bankruptcy with an average of 91% of the AUC metric. 

Keywords: Bankruptcy prediction, Boosting technique, Classification, Feature selection, Polish bankruptcy dataset, Python, Rule-based classification, Two-stage method, Weka, Wrapper methods.