CIGA+ : an Algorithm for Computing a Concise Set of Frequently closed Item Sets

Abstract

Since the output of a data mining task can be very large even for a reasonably small data set, the objective of the present paper is to describe an approach which reduces the data mining output and hence the execution time by approximating the set of frequent closed itemsets. More precisely, an algorithm called CIGA+ (Closed Itemset Generation and Approximation) is proposed and aims at partial or complete generation of frequent closed itemsets (FCIs) based on the construction and exploration of a dependency graph. The degree of approximation (eventually null) depends upon the value assigned to two parameter thresholds : cooccurrence frequency between two individual items and tolerance. Experimental analysis of our approach illustrates its cost-effectiveness and its potential for efficient association rule mining. Moreover, a comparative study with an existing and efficient algorithm for mining FCIs shows that CIGA+ has good performances even for large and dense data sets.