A novel privacy‐preserving scheme for collaborative frequent itemset mining across vertically partitioned data

Abstract

AbstractPrivacy preservation while undertaking collaborative data mining is a significant research problem. The vertically partitioned data model is an important data partition model and has varied applications. The vertically partitioned data model necessitates a non‐collusive scheme and an efficient scheme for the problem of privacy‐preserving distributed frequent itemset mining (PPDFIM). The current literature has schemes based on secure sum, set intersection cardinality and secure binary dot product (SBDP) for PPDFIM across vertically partitioned data. [m,m] Shamir's additive secret sharing has been proposed as a non‐collusive scheme for PPDFIM in a vertically partitioned setup that uses the secure sum sub‐protocol. However, such a scheme leads to information leakage in the distributed frequent itemset mining scenario and defeats the purpose of privacy preservation. We give a critique on the non‐collusive secret sharing‐based approaches when used for privacy preservation in frequent itemset mining in a vertically partitioned model. Further, we propose Du‐Atallah's efficient multiplication protocol for SBDP of two vectors for PPDFIM. We also propose an extension of the non‐collusive Du‐Atallah's SBDP protocol for a vertically partitioned setup to mine frequent itemsets for a multi‐party multi‐vector scenario. We show how such a collusion‐resistant scheme does not lead to loss of privacy and give the theoretical and empirical analysis therein. Further, we show that our proposed scheme is more efficient than the seminal public key‐based scheme proposed by Vaidya et al. in terms of the execution cost for a multi‐party scenario. Copyright © 2015 John Wiley & Sons, Ltd.