Abstract:

Discovery of outliers is to extract a few data objects with abnormal behavior patterns, which are more interesting than common patterns in some cases, from a large amount of data. It is of practical significance in IDS, credit fraud detection, etc. Data in these domains are usually high dimensional, particularly featured by their sparseness and decline properties. An algorithm that can obtain the outliers with high efficiency is proposed based on association analysis. Effectiveness of the algorithm is shown by theory analysis and experiment results.

Key words:

outlier, association analysis, closed frequent item-sets, k-relational neighboring area