Gradient Learning Approach for Variable Selection in Credit Scoring

摘要

The number of variables used for credit scoring can be quite large, and selecting the most relevant variables becomes an important topic. In this paper, we use gradient learning method for variable selection in credit scoring. The original method in the literature does not work on credit datasets because of the large sample size. To conquer this, we modify the algorithm by resampling data and voting effective variables. Compared with traditional variable selection methods, our approach can handle nonlinear models.