T. Alan Keahey, Data Mining Team, CIC-3
Tuesday, Aug. 27 10:00
CTI Conference Room
In this talk I will describe a tool for visualization of N dimensional points and clusters, which is used by the Data Mining Team for the detection of Medicare fraud. Data dimensions can be interactively mapped to the view volume axes to compose a "frame" of the data. Multiple frames can be rendered simultaneously in the view volume to facilitate visualization of more than 3 dimensions at a given time, or to compare the results from different analyses of the same data set.
The clustering algorithms which produce the ND points also generate probability density function estimates for each point and cluster. By making the transparency of the data points directly or non-linearly proportional to the probability, we can suppress the less interesting data points and visually isolate the low probability items.
Non-linear magnification techniques are also provided to enhance visualization. The entire view volume can be distorted to visually expand a single area or multiple areas. Bounded magnification can also be used to isolate magnification of individual clusters of points while maintaining a static global context.
There will be a live demo of the program.