Outlier detection in data streams using OLAP cubes

Abstract

Outlier detection is an important tool for many application areas. Often, data has some multidimensional structure so that it can be viewed as OLAP cubes. Exploiting this structure systematically helps to find outliers otherwise undetectable. In this paper, we propose an approach that treats streaming data as a series of OLAP cubes. We then use an offline calculated model of the cube’s expected behavior to find outliers in the data stream. Furthermore, we aggregate multiple outliers found concurrently at different cells of the cube to some user-defined level in the cube. We apply our method to network data to find attacks in the data stream to show its usefulness.