Outlier-based Data Association: Combining OLAP and Data Mining

Both data mining and OLAP are powerful decision support tools. However, people use them separately for years: OLAP systems concentrate on the efficiency of building OLAP cubes, and no statistical / data mining algorithms have been applied; on the other hand, statistical analysis are traditionally developed for two-way relational databases, and have not been generalized to the multi-dimensional OLAP data structure. Combining both OLAP and data mining may provide excellent solutions, and in this paper, we present such an example – an OLAP-outlier-based data association method. This method integrates both outlier detection concept in data mining and ideas from OLAP field. An outlier score function is defined over OLAP cells to measure the extremeness level of the cell, and when the outlier score is significant enough, we say the records contained in the cell are associated to each other. We apply our method to a real-world problem: linking criminal incidents, and compare our method with a similarity-based association algorithm. Result shows that this combination of OLAP and data mining provides a novel solution to the problem.