Privacy-Preserving Mining of Association Rules on Distributed Databases

Data mining techniques can extract hidden but useful information from large databases. Most efficient approaches for mining distributed databases suppose that all of the data at each site can be shared. However, source transaction databases usually include very sensitive information. In order to obtain an accurate mining result on distributed databases and to preserve the private data that is accessed, Kantarcioglu and Clifton proposed a scheme to mine association rules on horizontally partitioned data. This study proposes an Enhanced Kantarcioglu and Clifton Scheme’s (EKCS), which is a two-phase, privacy-preserving, distributed data mining scheme. It is based on the Kantarcioglu and Clifton’s Scheme (KCS) and reduces the quantities of global candidates that are encrypted and reduces the transmission load without raising the risk of itemsets leak in the first phase. Moreover, to increase the security against collusion in the second phase, this study proposes two protocols to be applied in the communication environment with or without a trusted authority, respectively. Key words: data mining, frequent itemset, privacy-preserving, securit