Incorporating support vector machines with multiple criteria decision making for financial crisis analysis

Feature selection is an essential pre-processing technique in data mining that eliminates redundant or unrepresentative attributes and improves the performance of classifiers. However, a classifier with different feature selection approaches results in diverse outcomes. Thus, determining how to integrate feature selection methods and yield an appropriate feature set is an issue worth further study. Based on ensemble learning, this investigation develops a SVMMCDM (support vector machines with multiple criteria decision making) model that employs various feature selection techniques as data preprocessing schemes and then uses SVM for financial crisis prediction. The study uses MCDM to determine the most suitable feature selection mechanism when many performance criteria are considered. After the feature selection mechanism has been determined, the study decomposes the SVM to obtain support vectors and predicted labels which are then fed into a decision tree to generate rules. The numerical results for the ex-ante and ex-post periods relative to the financial tsunami show that the proposed SVMMCDM model is an effective way to predict a financial crisis and can provide useful rules for decision makers.