The expected loss due to online fraud for the year 2008 is $4 billion, an increase of 11% on year 2007 loss of $3.6 billion. Although fraud detection has been continuously evolving, perpetrators continually develop new strategies to circumvent controls, and ongoing innovation in fraud auditing is required to keep fraud from exploding. Credit card fraud has been reported to have much larger implications, from organized crime and international narcotics trafficking to terrorist financing. The audit of credit card fraud is an ongoing 'arms-race' that requires constant innovation on the part of card issuers. Our research develops and tests a particular statistical innovation to attempt to better detect, and thus control and prosecute, credit card fraud.

In this study, we test models that account for misclassification error in credit card transactions, with a goal of assessing the performance of standard and modified binary choice models that include misclassification error parameters. We estimate models and the misclassification error parameters for two sample credit card transaction datasets. We found that the inclusion of omission error parameter in the modified model shifted the probability of fraud upwards, while as expected commission error was zero. The overall model adequacy measured by the percentage correct classification was similar for the standard and modified logit models.

This study is based on real-life credit card transactions dataset from an international credit card operation. This dataset has all credit card transactions during 13 months, from January 2006 to January 2007, of about 50 million transactions (49,858,600 transactions) on about one million (1,167,757 credit cards) credit cards from a single country.