Hybrid Probability-Based Ensembles for Bankruptcy Prediction

Bankruptcy prediction has attracted a lot of research interests as it is one of the major business topics. Both statistical approaches such as discriminant analysis, logit and probit models and computational intelligence techniques such as expert systems, artificial neural networks and support vector machines have been explored for this topic and most research compares the prediction performance via different techniques for a specific data set. However, there is no consistent result that one technique is consistently better than another. Different techniques have different advantages on different data sets and different feature selection approaches. Therefore, we divide the prediction performance into using different techniques for two parts: bankruptcy prediction and non­bankruptcy prediction. Based on analyzing the expected probability of both bankruptcy and non­bankruptcy predictions for a training set, we have built an ensemble of three well known classification techniques, i.e. the decision tree, the back propagation neural network and the support vector machine. This ensemble provides an approach which inherits advantages and avoids disadvantages of different classification techniques. In this paper we describe results which demonstrate that our expected probability­based ensemble outperforms other stacking ensembles based on a weighting or voting strategy.

@InProceedings{HCW07, 
 	 author =  {Hung, Chihli and Chen, Jing-Hong and Wermter, Stefan},  
 	 title = {Hybrid Probability-Based Ensembles for Bankruptcy Prediction}, 
 	 booktitle = {International Conference on Business and Information},
 	 editors = {},
 	 number = {},
 	 volume = {},
 	 pages = {},
 	 year = {2007},
 	 month = {Jul},
 	 publisher = {Springer},
 	 doi = {}, 
 }