In the world of financial transactions, rule-based heuristics are often employed to detect fraud, rather than to detect anomalies. The cat and mouse chase of financial institutions and fraudsters is an ongoing battle that has costed the global economy close to half a trillion dollars. There are some fundamental shortcomings of all rule based heuristics:
- Mistaking correlation for causation. In fact, machine learning and AI are thought of as solutions to overcome this shortcoming, but does that happen in reality?
- Rules are formulated based on past learnings and actions. If there are a set of rules, it’s very likely that a sophisticated fraudster already knows the semantics of most of these rules, hence ways to overcome or beat those rules
- Lack of a publicly auditable verification. The third drawback of the rule based or in fact any fraud management system is the proof of efficacy. Since fraud is such a delicate matter when it comes to compliance, no bank or financial institution will divulge the actual efficacy of a fraud management solution, because then they will have to divulge that they were actually subject to fraud
Given these, how does machine learning help in fraud detection?
Please download this free white paper to read more.