Multilevel Clustering of Induction Rules

Abstract

The tremendous size of data in nowadays world web invokes many data mining techniques. The recent emergence of some new data mining techniques provide also many interesting induction rules. So, it's important to process these induction rules in order to extract some new strong patterns called meta-rules. This work explores this concept by proposing a new support for induction rules clustering. Besides, a new clustering approach based on multilevel paradigm called multilevel clustering is developed for the purpose of treating large scale knowledge sets. The approach invokes k-means algorithm to cluster induction rules using new designed similarity measures. The developed module have been implemented in the core of the cognitive agent, in order to speed up its reasoning. This new architecture called Multilevel Miner Intelligent Agent (MMIA) is tested on four public benchmarks that contain 25000 rules, and compared to the classical one. As foreseeable, the multilevel clustering outperforms clearly the basic k-means algorithm on both the execution time and success rate criteria.