Mining of Multiobjective Non-redundant Association Rules in Data Streams

Abstract

Non-redundant association rule mining requires generation of both closed itemsets and their minimal generators. However, only a few researchers have addressed both the issues for data streams. Further, association rule mining is now considered as multiobjective problem where multiple measures like correlation coefficient, recall, comprehensibility, lift etc can be used for evaluating a rule. Discovery of multiobjective association rules in data streams has not been paid much attention. In this paper, we have proposed a 3-step algorithm for generation of multiobjective non-redundant association rules in data streams. In the first step, an online procedure generates closed itemsets incrementally using state of the art CLICI algorithm and stores the results in a lattice based synopsis. An offline component invokes the proposed genMG and genMAR procedures whenever required. Without generating candidates, genMG computes minimal generators of all closed itemsets stored in the synopsis. Next, genMAR generates multiobjective association rules using non-dominating sorting based on user specified interestingness measures that are computed using the synopsis. Experimental evaluation using synthetic and real life datasets demonstrates the efficiency and scalability of the proposed algorithm.