According to an empirical multiple discriminant analysis, one customer noted that some states parameters of an order were heavily involved in fraud. For example, it was found that some specific products or postal relay had a high rate of fraud made by its expertise from its own experience of several thousand orders.
A rigid system is not convenient for facing an evolving threat
After investigations to verify the observations of our client, we realized that some products might actually have higher rates of fraud than 80%: this means that about 10 orders containing this product, 8 were attempts of fraud. The observation on postal relay was also verified.
There was a need to dynamically identify and quantify the threat of risk statements on the data collected. Indeed, we have data on more than 180 000 orders and it is an important knowledge base on which it is possible to learn to continuously improve the performance of detection mechanisms.
We arrived at the conclusion that it was necessary that we implement a machine learning mechanism in our e-commerce Fraud Prevention Module NexuWeb Anti-Fraud E-commerce on products and addresses used.
An expert system by supervised learning to assist decision
An expert system is able to reproduce the cognitive mechanisms of an expert as a knowledge base and an inference engine. In our case, our knowledge base is composed of 180,000 orders labeled as legitimate order or fraud. The inference engine is a supervised learning algorithm for quantifying fraud rate for the two properties concerned: Products and delivery addresses.
Fraud evolves, changes and adapts
To improve detection performance against fraud, it is essential to reduce as much as possible human expertise to focus the work of the human agent on the decision. Faced with a very large amount of information as our 180 000 orders, it becomes complicated to require a precision of human expertise to deal with a threat that has evolved on information gathering time. Thus, systems integration experts on certain security controls provide real added value in the fight against a threat that is constantly evolving.