On the other hand, small data breach events (e.g, skimming) are happening on daily basis (a few hundred or thousand cards are skimmed here and there across millions of POS/ATM). They are less dramatic and wouldn’t make the headline. But their cumulative damage over a year is huge. That’s why we have developed Fraudulent Card Early Detection Solution.
Archive for the ‘fraud prevention rules’ Category
Some fraud detection rules are frivolous and they stop working soon after we create them. An example of possible frivolous rule looks like: bank card transactions at supermarket XYZ located at 100 Main Street, ABC City, are likely fraud. The rule could be created by analysts observing a number of fraud cases. However, it may be totally random that fraudsters conduct fraud transactions at supermarket XYZ. To create effective and stable rules, it is better to identify variables that capture the intrinsic fraudulent behavior patterns and are time-invariant and spatial-invariant.
When we set up fraud prevention rules, we should be aware of the complexity due to the interrelationships between variables. I use an example to illustrate the point. We can create two rules based on two variables:
Rule 1. Number of checks deposited in last 3 days
Rule 2. Number of checks deposited in last 5 days
Looking individually, both Rule 1 and Rule 2 are good. However, Rule2 does NOT detect many ADDITIONAL frauds because it overlaps with Rule 1. There is not much incremental value by including BOTH Rule1 and Rule 2 as part of the system.
So it is more effective to optimize rules holistically, i.e. , considering the multiple rules simultaneously, not individually. Each variable should provide new information and the overlap between variables should be small. After all, our goal is to optimize the system performance not the individual rule performance.