Community-Based Feature Selection for Credit Card Default Prediction

The prediction of credit card default is a critical issue in business and so has been attracting more and more attention. In this paper, we focus on the research of credit card default prediction in an unconventional way. We firstly study the consumption behavior of credit card holders, and uncover an interesting pattern that the features (each feature represents one dimension of the consumption behavior) cluster into different groups. With the aim of exploring the effect of the observed pattern on the task of credit card default prediction, we further propose a feature selection algorithm. Finally, we test the proposed algorithm and four existing feature selection algorithms on four prediction models over the real dataset of credit card consumption. Experimental results show that the proposed algorithm gives the overall superior performance and spends less running time in most cases; this demonstrates the potential application of the observed pattern.