DARCI: Distributed Association Rule Mining Utilizing Closed Itemsets

A distributed rule mining algorithm must minimize the communication cost to reduce the communication bandwidth use and to improve the scalability. There are a few distributed rule mining algorithms reported in the literature. In this paper we propose a new distributed association rule mining algorithm, called DARCI, which reduces the communication cost by up to 40% of the best known algorithm. The first step of mining global rules is to find globally frequent itemsets, and it involves the exchange of local supports of potentially frequent itemsets. DARCI reduces the communication cost by (1) sending frequent closed itemsets, which are supersets of locally frequent itemsets, instead of sending individual itemsets and (2) using a novel pruning technique to reduce the probability of having to send local supports.