To Maintain The Privacy Of Data Mining Research

Data mining technology has emerged as a means for extracting interesting patterns or knowledge from large quantities of data, and is becoming widely used in the fields such as scientific research, medical research and business. However, data mining, with its nature to efficiently discover valuable, non-obvious patterns or rules from massive data, is particularly vulnerable to misuse. Privacy preserving data mining considers how to collaborate to obtain globally valid data mining results without revealing any unnecessary information of the sites. That is a significant direction for future data mining research.This paper concentrates on the issue of privacy preserving data mining. Specifically, aiming at two widely used algorithms in data mining, Naive Bayesian Classifier and Boolean Association Apriori algorithm. We have brought forward two corresponding protocols incorporating privacy concerns. We have used secure multi-party computation protocols and tools to get the solutions. Main work includes:1) We present an oblivious polynomial evaluation protocol. The oblivious polynomial evaluation protocol will be used many times in our privacy preserving naive bayesian classifier,so its efficiency is important to the solution. By transforming many invocations of OT 21 to oneinvocation of OTN1, we present a high-efficient oblivious polynomial evaluation protocol.2) We construct a secure division protocol. The starting point for the solution is the Taylorseries of the 1/x, and then by using a single private polynomial evaluation protocol we can get the solution.3) We construct the privacy preserving Naive Bayesian Classifier. By constructing two secure posterior probability evaluation protocols to deal with discrete and numeric, or categorical and continuous attributes respectively, we attain the naive bayesian classifier without preamble. We prove that our protocol is accurate and private, and is highly efficient.4) We present a protocol to solve the open problem brought forward by Vaidya, i.e, how to construct privacy preserving boolean association rule mining protocol in vertically partitioned data of multiple parties. The protocol presented discovers globally frequent itemsets with minimum support levels;'restricting the shares of the individual transaction across the sites and protect either site from revealing any unnecessary information farthest. And we also present a multi-party security scalar product protocol.