Frequent Closed Itemset Mining with Privacy Preserving for Distributed Databases

Abstract

In the present paper we introduce closed item sets into frequent item set mining from horizontally-partitioned transaction databases with preserving privacy. Closed item sets were originally from the research area of Formal Concept Analysis, and it is shown that even if results of frequent item set mining are restricted to closed item sets, all frequent item sets can be recovered from the results. This property suggests that using closed item sets would contribute to decreasing the cost of communication among distributed sites where a piece of horizontally-partitioned database is stored. We present a mining procedure revising and amalgamating two previous works: one is for mining closed item sets from horizontally-partitioned databases, and the other is for privacy preserving mining of item sets from such databases. We analyze the procedure on both of the viewpoint of communication cost and that of security. We also show results of some experimental practice of applying the procedure to a well-known dataset.