Generalist Leaders, Specialized Subordinates - An Ensemble Learning Approach for Bankruptcy Prediction

Joel Quek, Ke-Wei Huang

Abstract: Description Bankruptcies can have serious implications for regulators, investors and the economy due to increasing systemic risk in modern times. The state-of-the-art machine learning forecasting solution may involve using ensemble approaches such as stacking and other weight averaging methods. However, most of the ensemble methods focus on building one single global meta classifier by which we do not know much about how this model handles heterogeneity of firms in different industries. We propose a new algorithm that can better address the heterogenous nature of firm properties for bankruptcy prediction. In addition, our ensemble model is designed to work with any existing classification algorithm, to better handle the low proportion of bankruptcy cases, and can incorporate human-specified rules from domain knowledge. Our preliminary experiment results indicate that our proposed model can indeed improve the prediction performance on five commonly used classification algorithms.

External IDs:dblp:conf/icis/QuekH21