HighPU: a high privacy-utility approach to mining frequent itemset with differential privacy

In the field of data mining, frequent itemset mining (FIM) is a popular technique for analysing transaction datasets and establishing the foundation of association rules. Publishing frequent itemsets, however presents privacy challenges. Differential privacy provides strong privacy assurance to users. In this paper, we study the problem of mining frequent itemsets under the rigorous differential privacy model. We propose an approach, called HighPU, which achieves both high data utility and high degree of privacy in FIM. HighPU begins by truncating transactions over the original dataset. Then HighPU directly searches for maximal frequent itemsets. And we use a consistent approach to improve the accuracy of the results. Extensive experiments using several real datasets illustrate that HighPU significantly outperforms the current state of the art.