A sliding window based meta-majority of voting ensemble for credit risk assessment

In this paper an ensemble of classifiers for credit risk assessment is proposed. A sliding window of samples with pre-specified size is adopted to train each individual classifier of ensemble in a logical ring structure. The completion of one cycle of a logical ring structure is treated as a pass. During every pass by voting mechanism classifier with higher accuracy is maintained. The final accuracy of the ensemble method is determined after completion of all cycles with a meta-voting. We have evaluated the performance of our method on two publicly available credit databases and compared it with two benchmark ensemblers such as Bagging and Boosting. Type-I and Type-II errors of this method suggest the financial institutions to assess their credit risk accurately and make them healthy.