An ensemble-based model for two-class imbalanced financial problem

This study proposes an ensemble-based model (EBM) for the two-class imbalanced classification problem by joining together the support vector machine (SVM), multiple feature selection combination, back-propagation neural network (BPNN) ensemble, and rough set theory (RST). To improve the significance of the rare and specific region belonging to the minority class in the decision region, we take the SVM as a pre-processor to balance the training dataset and use multiple feature selection combination grounded on ensemble learning in order to determine the most representative features from the re-sized dataset. The representative features are then fed into the BPNN ensemble to construct an effective financial pre-warning mechanism. Lacking comprehensibility and readability is one of the fatal weaknesses of an ensemble classifier and it impedes its real-life application. Thus, the study executes RST to extract knowledge from the BPNN ensemble for decision makers to make suitable judgments. Decision makers can take the decision rules as a roadmap to modify a firm's capital structure so as to survive in an extremely turbulent financial market. Empirical results reveal that the introduced EBM's prediction accuracy is very promising in financial risk mining, relative to other detection approaches in this study.