An evaluation of support vector machines in consumer credit analysis

Abstract

This thesis examines a support vector machine approach for determining consumer credit. The support vector machine using a radial basis function (RBF) kernel is compared to a previous implementation of a decision tree machine learning model. The dataset used for evaluation was provided by a large bank and includes relevant consumer-level data, including transactions and credit-bureau data. The results suggest that a support vector machine offers similar performance to decision trees, but the parameters specifying the soft-margin constraint and the inverse-width used in the RBF kernel could significantly affect its performance.