Discriminant analysis and neural network methodologies were applied to the problem of identifying illegal sales transactions. The researchers independently developed models using data provided by a cr^it card company. A series of measures were developed and used to construct the models. The final results were that the discriminant analysis model recognized 32.3% of the fraudulent activity, while the neural network approach found 28.9%. With only 11.6% of the transactions in common, the combination of the two models identified 49.6%i. In order to provide a real time monitoring program, the models were simplified yielding a capture rate of approximately 42%.
Richardson, Robert J.
"Monitoring Sale Transactions for Illegal Activity,"
Communications of the IIMA: Vol. 6
, Article 10.
Available at: https://scholarworks.lib.csusb.edu/ciima/vol6/iss1/10