Mining of Rare Itemsets in Distributed Environment

The mining of rare itemsets involves finding rarely occurring items. It is difficult to mine rare itemsets with a single minimum support (minsup) constraint because low minsup can result in generating too many rules in which some of them can be uninteresting [3]. In the literature [4, 5], "multiple minsup framework" was proposed to efficiently discover rare itemsets. However, that model still extracts uninteresting rules if the items' frequencies in a dataset vary widely. In this paper, we are using the notion of "item-to-pattern difference" and multiple minsup based FP-growth-like approach proposed in [6] to efficiently discover rare itemsets in the distributed environment. To discover global rare itemsets in distributed environment, information regarding itemsets of local sites is collected in the form of MIS-tree at one site; that is, each site sends its local MIS-tree to a single site where a global MIS-tree will be constructed from all the MIS-trees received from all the sites. This global MIS-tree is mined to generate global rare itemsets. Experimental results show that this approach is efficient in terms of communication bandwidth consumed.