Outlier and Anomaly Detection in fraud scenarios
When online fraud is committed, fraudsters take advantage of gaps allowing them to unjustifiably enrich themselves. However, when fighting fraud, companies face a dilemma, given that no system is perfect: in ecommerce, on the one hand, fraud and its related losses should be reduced; on the other hand, users neither want to be accused of fraud nor treated like criminals. In other areas, such as health, the problems associated with data abuse and security leaks could even result in more severe damage than purely financial matters.
Yet, companies’ practical implementations of fraud investigation processes rarely meet scientific standards. Fraud prevention companies offer diverse products on the market, but neither their effectivity nor their efficiency has been verified scientifically until now. Dealing with these issues, the first step should be a clean definition of what constitutes fraud and how it can be measured. Afterwards, models can be built and tested according to scientific standards of evaluation. On this basis, a careful risk analysis can be conducted in order to weigh pros and cons of pursuing individual suspicious cases.
Companies spend millions to protect themselves from fraudulent activities. One of the most interesting aspects is the fact that fighting fraud is a social interaction that needs constant supervision and improvement: a never ending race between criminals and investigators, in which both actors adjust for the actions of the other.