Association rules exploration through clustering

Abstract

Association, one of the data mining techniques, identifies all the intrinsic relationships in database. However, this characteristic, which can be advantageous on the one hand, generates a large number of patterns. Thus, the post-processing of the discovered rules becomes an important topic, since it is necessary to validate the rules. To do so, many approaches have been used, as the clustering approach. The aim in applying clustering in the post-processing is to generate groups of rules in order to facilitate the exploration of the obtained knowledge. However, another way to facilitate the exploration is avoiding the generation of a huge amount of rules using clustering in the pre-processing. In this case data are initially grouped and from each group the association rules extraction is realized.Based on the exposed arguments, the objectives of this project are: (i) explore the use of clustering as a way of post-processing association rules; (ii) explore the obtainment of association rules through clustering in the pre-processing; (iii) compare (i) and (ii) in order to identify which of the approaches present the best results considering the association rules exploration. Thus, it is expected with this project to verify if the clustering technique supports the exploration of association sets and, in affirmative case, identify: (i) which clustering algorithms/measures generates the best results; (ii) which is the step, pre and/or post-processing, more adequate to use clustering. That way, this project will contribute with the data mining area making viable to researches a better understanding of the clustering technique when used in association rules exploration. (AU)