| Abstract |
During the last year I worked on 3 projects aimed at detecting fraud in various sorts of data:
- health insurance
- US Crop Insurance Program
- electronic payments
In my talk I will discuss these cases in some detail, focusing mainly on efficient data structures that make the detection process very efficient. In particular, I will discuss static hash tables and multi-linked rotating buffers that I used in a prototype system that monitors, in real time, 1 billion (1000.000.000) of the most recent transactions (using the sliding window model). Additionally, I will share some experience with using kd-trees for analyzing spatial data from the US Department of Agriculture. |