By Published on .

Most Popular

The person on the other end of the line, a rep from my credit card company, wanted to verify a few recent purchases on my card. Did I buy a $1,500 vacation package lately? No. How about a last-minute round-trip ticket to Martinique? Uh-uh. Or $500 worth of women's sportswear? I wish. Within a week, the rep informed me, my account balance had soared from $64 to nearly $5,000. It sure wasn't me on the shopping spree, even though my credit card sat snugly in my wallet. Someone had stolen my account number and was ringing up purchases, mostly over the phone. At that very moment, they were probably sipping a cocktail on a beach in Martinique.

A neural network discovered the unusual spike in my account, and that prompted a subsequent phone call from the company's fraud department. Data mining has been around for years in the credit card business, as well as in other industries like health care, retail, and telecommunications. Still, credit card fraud remains a $1.5 billion problem worldwide, according to The Nilson Report, a trade publication that tracks the consumer payments sector. E-commerce isn't helping matters: Internet Fraud Watch, operated by the National Consumers League, reports that Web-related complaints increased 600 percent from 1997 to 1998. Today, experts are looking for new data mining applications to more accurately pinpoint fishy behavior, online and off.

Textual data mining may be one potential extension. It's already at work in other areas-type "Avis" into Infoseek's search engine on the Web, for example, and up pops an ad for the rental car company with the search results. There are applications in fraud detection as well, says Wesley Wilhelm of San Diego-based HNC Software, a provider of fraud detection technology to nine out of the top ten Visa and MasterCard credit card issuers in the United States. A merchant could mine a customer's instructions, such as "leave package on the porch," on an order form, he suggests. Neural networks would translate the raw text into a numerical algorithm and score the likelihood of fraud. Of course, there can be false positives. A customer might request that her order be delivered to a next-door neighbor simply because she's going on vacation. A follow-up phone call can save the order, and even boost loyalty to the brand. In my case, the fraudulent purchases never appeared on my statement, and a new card was issued to me immediately. Did the experience affect my loyalty to the credit card company? Let's just say I haven't switched yet. -JL

In this article: