A Clustering Based Classifier Ensemble Approach to Corporate Bankruptcy Prediction

Corporate bankruptcy prediction is an important research direction in finance. Building a robust prediction scheme for bankruptcy can be beneficial to several stakeholders, including management organizations, government and stockholders. Ensemble learning is a well-known technique to improve the predictive performance of classification algorithms by decreasing the generalization error and enhancing the classification accuracy. It has been a well-established technique in bankruptcy prediction to enhance the predictive performance. Diversity plays an essential role in constructing robust ensemble classification schemes. In this paper, a clustering based classifier ensemble approach is presented for corporate bankruptcy prediction. In this scheme, k-means algorithm is utilized to obtain diversified training subsets. Based on the subsets, each base learning algorithms are trained and the predictions of base learning algorithms are combined by a majority voting scheme. In the empirical analysis, four classification algorithms (namely, C4.5 algorithm, k-nearest neighbour algorithm, support vector machines and logistic regression) and three ensemble learning methods (Bagging, AdaBoost and Random Subspace) are evaluated.