Hybrid Intelligent Method for Association Rules Mining Using Multiple Strategies

Abstract

Association rules mining has attracted a lot of attention in the data mining community. It aims to extract the interesting rules from any given transactional database. This paper deals with association rules mining algorithms for very large databases and especially for those existing on the web. The numerous polynomial exact algorithms already proposed in literature processed the data sets of a medium-size in an efficient way. However, they are not capable of handling the huge amount of data in the web context where the response time must be very short. Moreover, the bio-inspired methods have proved to be paramount for the association rules mining problem. In this work, a new association rules mining algorithm based on an improved version of Bees Swarm Optimization and Tabu Search algorithms is proposed. BSO is chosen for its remarkable diversification process while tabu search for its efficient intensification strategy. To make the idea simpler, BSO will browse the search space in such a way to cover most of its regions and the local exploration of each bee is computed by tabu search. Also, the neighborhood search and three strategies for calculating search area are developed. The suggested strategies called (modulo, next, syntactic) are implemented and demonstrated using various data sets of different sizes. Experimental results reveal that the authors' approach in terms of the fitness criterion and the CPU time improves the ones that already exist.