Because most fraud detection systems tend to be rules-based, one very surprising but common limitation is that most are not good at identifying new types of fraud and abuse. Out-of-the-box standard data mining approaches are not much better either because they are based on past observations (which don't usually exist) or because they use undirected approaches which suffer from high false positive rates. While traditional analytics methods are not designed to look for rare or unknown events, Dr. Ian J. Scott will present a real-life solution to this problem based on techniques used to discover new particles in high energy physics. Applied to analytics, the approach involves learning from your own data using an unbiased and automatic approach to fraud detection.

The Angoss Problem Solver series presents relevant and engaging topics to illustrate how data mining tools and techniques can be applied in creative ways to solve real-world business problems.