Multiple extreme learning machines for a two-class imbalance corporate life cycle prediction

• Published in: Knowledge-based systems 2013-02, Vol.39, p.214-223
• Main Authors: Lin, Sin-Jin, Chang, Chingho, Hsu, Ming-Fu
• Format: Article
• Language: English
• Subjects: Corporate life cycle; Decision making; Ensemble learning; Extreme learning machine; Imbalanced dataset; Knowledge generation
• ISSN: 0950-7051; 1872-7409
• DOI: 10.1016/j.knosys.2012.11.003

Pre-warning of whether a corporate will fall into a decline stage in the near future is an emerging issue in financial management. Improper decision-making by firms incurs a higher possibility to cause financial crisis (distress) and deteriorates the soundness of financial markets. The aim of this study is to establish a novel prediction mechanism based on combining the sampling technique (synthetic minority over-sampling technique; SMOTE), feature selection ensemble (original, intersection, and union), extreme learning machine (ELM) ensemble and decision tree (DT). The proposed model – namely, the multiple extreme learning machines (MELMs) – shows promising performance under numerous assessing criteria, but one critical defect of the ensemble classifier is that it lacks comprehensibility. Thus, we perform a DT as the knowledge generator to extract the inherent information from the ensemble mechanism. This knowledge visualized process can assist decision makers in efficiently allocating limited financial resources and to help firms survive in an extremely competitive environment.