Home // ICCGI 2012, The Seventh International Multi-Conference on Computing in the Global Information Technology // View article

Minimising Expected Misclassification Cost when using Support Vector Machines for Credit Scoring

Authors:
Terry Harris
Curtis Gittens

Keywords: Credit Scoring; Decision Support Systems; Expected Misclassification Cost ; Support Vector Machines

Abstract:
With the gradual relaxation of credit around the world, the cost of losses experienced when extending credit is expected to become increasingly important to financial institutions. In this paper, we offer theoretical and empirical evidence to support the argument that the minimisation of this cost should be the primary objective when developing classification models for credit scoring. This cost can be referred to as the Expected Misclassification Cost. In addition, we present and test a system that builds models to minimise this cost when given varying values for its components. Moreover, we show that using differing values for the components of Expected Misclassification Cost can result in improved performance, in terms of Type I or Type II accuracy, when Expected Misclassification Cost is used as the prime evaluation metric by a support vector machine.

Pages: 225 to 231

Copyright: Copyright (c) IARIA, 2012

Publication date: June 24, 2012

Published in: conference

ISSN: 2308-4529

ISBN: 978-1-61208-202-8

Location: Venice, Italy

Dates: from June 24, 2012 to June 29, 2012