Feature Subset Selection using Adaptive Differential Evolution: An Application to Banking

In this paper, we developed a feature subset selection method by employing Adaptive Differential Evolution as a wrapper. The proposed wrapper utilizes four independent classifiers namely Logistic Regression, Probabilistic Neural Network, Naive Bayes and Support Vector Machine. We employed the Matthews Correlation Coefficient (MCC) as the fitness function or evaluation measure. In order to demonstrate the efficacy of the proposed method, we tested on three datasets, of which two are related to credit scoring and one is related to financial statement fraud. Our proposed method yielded better results than other standard methods in the literature as well as Differential Evolution. We also performed a statistical significance test i.e. t-test at 1% level of significance, which infers that some of the proposed wrappers are statistically one and the same.