Association Rules Mining Based on the Discriminative Concept Lattice

Abstract

As a fundamental data mining task, frequent pattern mining has widespread applications in many different domains. Concept lattice is a very useful formal analysis tool. There is a one-to-one correspondence between concept intensions and frequent itemsets. Abundant literature has been conducted in-depth research in mining frequent itemsets and association rules based on the concept lattice. Most of them, however, did not take into account the differences of attributes when building concept lattice. In this paper, we present a new concept called discriminative attributes (discriminative intension). Every attributes has different discriminative power (DISP). In the process of building concept lattice, we can remove the attributes of low discriminative power, then speed up the step of constructing lattice. Furthermore, there has been a new method to calculate the DISP and reset the value under some conditions, but scan the database on each layer, which reduce the times of scanning the database, then decrease the time to generate association rules.