Data Mining for Individual Consumer Credit Default Prediction under E-commence Context: A Comparative Study

This paper explored the relative effectiveness of alternative classifiers to estimate how likely an applicant is to default in individual consumer credit services offered by e-commerce platform or online payment company. Specifically, our work thus contributes to the following research questions: (i) What features should be introduced in the new context of e-commence (e.g. social features)? Which features plays important roles in credit scoring? (ii) How to tuning classification algorithms in an efficient way to avoid model inefficiency? (iii) Do ensemble classifiers real improve classification ability? Data mining methods were adopted in the effort to answer these questions. The testing results indicated that extreme gradient boosting, a novel ensemble classifier, seems to be very adequate to be used for credit scoring of its good performance under imbalanced credit scoring sample. In addition, we also conducted feature importance analysis and enhanced the interpretability of credit scoring model