A new multi-criteria convex quadratic programming model for credit analysis

Abstract

Mathematical programming based methods have been applied to credit risk analysis and have proven to be powerful tools. One challenging issue in mathematical programming is the computation complexity in finding optimal solutions. To overcome this difficulty, this paper proposes a Multi-criteria Convex Quadratic Programming model (MCCQP). Instead of looking for the global optimal solution, the proposed model only needs to solve a set of linear equations. We test the model using three credit risk analysis datasets and compare MCCQP results with four well-known classification methods: LDA, Decision Tree, SVMLight, and LibSVM. The experimental results indicate that the proposed MCCQP model achieves as good as or even better classification accuracies than other methods. © Springer-Verlag Berlin Heidelberg 2006.