A scalable and efficient privacy preserving global itemset support approximation using bloom filters

Abstract

Several secure distributed data mining methods have been proposed in the literature that are based on privacy preserving set operation mechanisms. However, they are limited in the scalability of both the size and the number of data owners (sources). Most of these techniques are primarily designed to work with two data owners and extensions to handle multiple owners are either expensive or infeasible. In addition, for large datasets, they incur substantial communication/computation overhead due to the use of cryptographic techniques. In this paper, we propose a scalable privacy-preserving protocol that approximates global itemset support, without employing any cryptographic mechanism. We also present some emperical results to demonstrate the effectiveness of our approach. © 2014 IFIP International Federation for Information Processing.