Fast Computation of 2-Dimensional Depth Contours

"Oneperson's noise is another person's signal." For manyapplications, including the detection of credit card frauds and the monitoringof criminal activities in electronic commerce,an important knowledgediscovery problemis the detection of exceptional/outlying events. In computational statistics, a depth-based approach detects outlying data points in a 2-D dataset by, basedon somedefinition of depth, organizingthe data points in layers, with the expectation that shallowlayers are morelikely to contain outlying points than are the deeplayers. One robust notion of depth, called depth contours, was introduced by Tukey. ISODEPTH,developed by Ruts and Rousseeuw,is an algorithm that computes 2-Ddepth contours. In this paper, wegive a fast algorithm, FDC, which computes the first k 2-D depth contours by restricting the computationto a small selected subset of data points, instead of examiningall data points. Consequently, FDCscales up much better than ISODEPTH.Also, while ...

• Type: Conference
• Year: 1998
• Where: KDD
• Authors: Theodore Johnson, Ivy Kwok, Raymond T. Ng
• Keywords: Data Mining, Data Points, Depth Contours, KDD 1998, Organizingthe Data Points