8.22 The Hybrid Solution


8.22 The Hybrid Solution

Typically, a fraud-detection system employs a mixture of human domain knowledge and data mining technologies, such as machine-learning and neural-networks. Empirical studies have shown that hybrid systems, combining human insight and machine-learning rules with neural-network models, are the best for detecting fraud. They combine the better of two worlds and represent the strength of intuitive human instincts and powerful pattern-recognition technologies. It is the experienced hunter aided by the relentless hunting dog, searching for data clues and potential criminal activity.

For example, a fraud-detection specialist who is knowledgeable about what type of products tend to attract criminals would know that computer hardware is an item that can easily be unloaded at auction sites. However, a machine-learning algorithm can detect the range of transactions (3–5) and price intervals, out of thousands of transactions, that signal a potential fraudulent transaction. A data mining analysis can segment, in real time, several transactional and consumer variables, enabling the following type of real-time alert to be generated:

      IF card expiration is 2 years from this month      AND number of purchases = 3 - 5      AND Standard Industry Code (SIC) of product = hardware      AND average price of purchases < $254 dollars      AND median rent = $425-548      THEN fraud probability score = 88% 

One of the most difficult challenges to fraud detection is finding a new signature or pattern. For data mining to be used in the deterrence of fraud, samples of transactions that turn out to be fraudulent have to be used to train the system to extract either a scorecard, in the mode of a formula of weights from a neural network, or an array of rules created from a data mining analysis of fraudulent and nonfraudulent transactions extracted via a machine-learning algorithm such as C5.0 or a segmentation engine such as CART.

In addition to the use of data mining, coupled with a trained investigator's domain expertise, there are several deterrence techniques that can be used to detect and deter on-line credit-card fraud. For example, companies such as InfoSplit can provide the geo location of the IP address of a purchaser, which a merchant can compare against the shipping location to investigate all mismatches.

Other clickstream information, such as the time of day, browser type, operating system, and other on-line behavior data, such as the referring sites, can be included in the data mining process for developing a new set of models for detecting fraud. Unquestionably, the biggest obstacle to the future of successful e-businesses today is credit-card fraud, both for e-merchants and for consumers subject to auction fraud.

According to Meridien Research, without any technological investments in fraud detection and prevention, worldwide credit-card fraud will represent $15.5 billion in losses by 2005. However, if merchants adopt data mining technology now to help screen credit-card orders prior to processing, the widespread use of this technology is predicted to cut overall losses by two-thirds to $5.7 billion in 2005. Criminal analysts commonly work with statistics to determine trends and levels of threat to individuals and property, so it is not unrealistic to see how data mining, a technology rooted in the use of machine-learning and statistical algorithms, can be used to stem the flow of fraud in this emerging on-line marketplace.

The battle against on-line fraud is an unending one; e-merchants and Web auctioneers must be shrewd and use data mining technologies and experts to stay ahead of the scams of criminals, such as the gang, that had the audacity to invest in its own ATM machine, stock it with cash, and then wheeled it into a busy shopping mall for the purpose of "skimming" credit- and debit-card numbers, along with their associated PIN codes. This is a true story, and one where the perpetrators got away clean. Similar scams are rapidly evolving every day on the Web, where criminals are silently rolling out their virtual ATM machines in hopes of defrauding merchants of their money and consumers of their identities.




Investigative Data Mining for Security and Criminal Detection
Investigative Data Mining for Security and Criminal Detection
ISBN: 0750676132
EAN: 2147483647
Year: 2005
Pages: 232
Authors: Jesus Mena

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net