Finding disjoint clusters in a categorical data space

Abstract

In This paper we provide a prototype of method for segment a high dimensional categorical data using frequent patterns. The frequent patterns are mined using a conventional frequent pattern mining algorithm according to a predefined support threshold. In addition, we restrict the frequent patterns length to a predefined low value in order to ensure the understandability of the results. Associations between the frequent patterns are discovered in order to reveal containment and overlap between them. Segments are iteratively defined as the largest region of data space covered by several frequent patterns. The illustrative example shows promising results in term of the quality of the resulted segments and the understandability.