Group for Research in Decision Analysis

G-2019-06

Performance of the mathematical programming approach in credit scoring

, , , and

Operations Research (OR) has a very important role to play in credit scoring for building models that can help the lending organization to make a good decision about accepting or rejecting a new applicant (borrower). Many standard statistical and machine learning techniques are used in the literature, for example, Linear Discriminant Analysis, Linear Regression, Hidden Markov Model, Support Vector Machine and Artificial Neural Network. In this paper, we propose a new approach for credit scoring. The idea is to combine statistics and OR modeling to solve this problem. An originality of our approach is that it proceeds by three steps. First, we use the cross validation method for separating the testing part from the training part. In the second step, a mathematical programing approach is used in discriminant analysis. In the last step, we use resampling techniques (like Jackknife and Bootstrap procedures) for estimating the parameters for the discriminant mathematical programming model. The performance of the proposed approach is validated on two Australian and German public credit datasets.

, 13 pages