P. Chan and S. Stolfo. The effects of training class distributions on performance using cost models. Submitted to 15th Intl. Conf. Machine Learning, 1998.  Home/Search   Document Not in Database   Summary   Related Articles

....the credit card fraud detection task in Section 2. Section 3 examines the effects of training class distributions on the performance. Section 4 discusses our multiclassifier meta learning approach. For completeness, part of the exposition in this article also appears in a companion paper [8]. Section 5 summaries our results and directions. 2 Credit Card Fraud Detection When banks lose money because of credit card fraud, card holders partially (possibly entirely) pay for the loss through higher interest rates, higher membership fees, and reduced benefits. Hence, it is both the banks ....

....the bank can afford to send larger number of transactions for investigation. That is, the bank can tolerate more false alarms (a higher false positive rate) and aim for fewer misses (a lower false negative rate) which can be achieved by a larger percentage of fraudulent transactions (positive s) [8]. Conversely, if the overhead is larger, the bank should aim for fewer false alarms (a lower FP rate) and tolerate more misses (a higher FN rate) which can be obtained by a smaller percentage of positive s. Note that, at some point, the overhead can be large enough making fraud detection ....

P. Chan and S. Stolfo. The effects of training class distributions on performance using cost models. Submitted to 15th Intl. Conf. Machine Learning, 1998.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC