A selective ensemble based on expected probabilities for bankruptcy prediction

Bankruptcy prediction is one of the major business classification topics. Both statistical approaches and artificial intelligence techniques have been explored for this topic. Most researchers compare the prediction performance using different techniques for a specific data set. However, there are no consistent results to show that one technique is better than the other. Different techniques have different advantages and disadvantages on different data sets. Recent studies suggest combining multiple classifiers may have a better performance. However, such an ensemble is usually not only to inherit advantages from the different classifiers but also suffers from disadvantages of those classifiers. In this paper, we propose a selective ensemble of three classifiers, i.e. the decision tree, the back propagation neural network and the support vector machine. Based on the expected probabilities of both bankruptcy and non-bankruptcy, this ensemble provides an approach which inherits advantages and avoids disadvantages of different classification techniques. Consequently, our selective ensemble performs better than other weighting or voting ensembles for bankruptcy prediction.