Authors: Youcef Djenouri; Habiba Drias; Ahcene Bendjoudi
Addresses: LRIA, USTHB: University of Algiers, BP 32 El Alia Bab Ezzouar, Algers, Algeria ' LRIA, USTHB: University of Algiers, BP 32 El Alia Bab Ezzouar, Algers, Algeria ' LRIA, USTHB: University of Algiers, BP 32 El Alia Bab Ezzouar, Algers, Algeria; CERIST Research Center, 16030, Ben-Aknoun, Algiers, Algeria
Abstract: The efficiency of existing association rules mining algorithms afford large number of delivered rules that the user can not exploit them easily. Consequently, thinking about another mining of these generated rules becomes essential task. For this, the present paper explores metarules extraction in order to prune the irrelevant rules. It first focuses on clustering association rules for large datasets. This allows the user better organising and interpreting the rules. To more go down in our mining, different dependencies between rules of the same cluster are extracted using meta-rules algorithm. Then, pruning algorithm uses these dependencies to delete the deductive rules and keep just the representative rules for each cluster. The proposed approach is tested on different experiments including clustering, meta-rules and pruning steps. The result is very promising in terms of the number of returned rules and their quality.
Keywords: data mining; association rules; rule pruning; swarm intelligence; optimisation; knowledge mining; metarules; metarule extraction; irrelevant rules; clustering.
International Journal of Business Intelligence and Data Mining, 2014 Vol.9 No.2, pp.112 - 144
Received: 27 Jan 2014
Accepted: 18 Mar 2014
Published online: 24 Oct 2014 *