Predicting Corporate Failure Using Ensemble Extreme Learning Machine

Corporate failure prediction has become a major topic in the accounting and finance literature. Effective prediction models are essential for banks and financial institutions to solve financial decision-making problems. In general, artificial intelligence and machine learning techniques have been mainly employed to develop corporate failure models due to their prediction superiority in comparison to the traditional statistical method. Extreme learning machine is a newly developed artificial intelligence technique with an extremely fast learning speed. Nonetheless, its performance instability may be a major constraint for its practical application. The literature documents that the ensemble is one of the widely used methods to improve the generalization performance of weak classifiers. Therefore, we propose in this study an ensemble of extreme learning machine for improving the prediction performance on corporate failure task. In particular, we compare four benchmark ensemble methods (multiple classifiers, bagging, boosting, and random subspace) to evaluate which is best suited for extreme learning machine. Experimental results on French firms indicated that bagged and boosted extreme learning machine showed the best-improved performance.