Privacy-preserving Association Rules Mining

With the rapidly developing of information technologies, especially of network technology, data storage technology and high capability processor technology, it is more and more convenient to collect, manage and analyse massive data. Knowledge Discovery and Data Mining have more positive effect in deep application, but every coin has two sides and data mining is not exceptional. Information security and privacy preserving bring up with the development of data mining. So, how to mine useful and exact information on the privacy preserving is becoming hotspot in data mining field at present.Firstly, the current privacy preserving data mining algorithms are introduced and analysed, as data distributing method, data modifying method etc. Secondly, three privacy-preserving association rules mining methods MASK, RRPH and PARD are important to introduce. Finallly, a new privacy-preserving association rules mining method (PRRPM) is proposed in this paper, which is a novel partial randomized response method based on probability matrix.PRRPM method chooses different data transition strategies to find frequent 1-itemsets and frequent k-itemsets (k>1) in order to mine association rules accurately and efficiently while preserving privacy. When mining frequent 1-itemsets, a new method is presented to recove the support of frequent 1-itemsets in original data set to find all frequent 1-itemsets after transforming every attribute partially with 1-itemsets transition probability matrix; when mining frequent k-itemsets (k>1), a same method is presented to recover the support of candidate frequent k-itemsets (k>1) in original data set to find all frequent k-itemsets (k>1) after transforming every candidate frequent k-itemsets (k>1) partially with multi-itemset transition probability matrix.Theory analysis and experiments show that the PRRPM method is effectively and better than those of MASK and RRPH in the privacy, accuracy and complexity.