Genetic Algorithms for the Optimization of Support Vector Machines in Credit Risk Rating

The assessment of credit risk usually involves the development of rating models that classify credit applicants (firms or individuals) into predefined risk groups. A plethora of methodologies have been proposed to develop such rating models. Among them support vector machines (SVMs) have rapidly evolved in statistical learning theory as new modeling technique for developing classification models. However, their application requires the specification of several parameters. This paper proposes the use of genetic algorithms for the determination of optimal parameters for SVM models developed for credit risk assessment. The proposed methodology is applied to three data sets related with the development of credit scoring systems and is compared with discriminant analysis and logistic regression.