Principal Component Analysis and ReliefF Cascaded with Decision Tree for Credit Scoring

The objective of this paper is to propose a credit scoring approval model using a feature selection technique performed by Principal Component Analysis (PCA) and ReliefF algorithm followed by a decision tree classifier. As a reference classifier, we have chosen Support Vector Machine (SVM). The performance of our proposed model has been tested using the German credit dataset. The experimental results of the proposed signal processing cascade for the credit scoring lead to the best accuracy of 91.67%, while classifiers without feature selection show the best accuracy of only 75.35%. On the other side, using the same combination of feature selection (PCA and ReliefF) but cascaded with SVM classifier, one has obtained an accuracy of only 85.15%. The experimental results confirm the accuracy of the proposed model, and at the same time they show the importance of feature selection and its optimization for credit scoring decision systems.