Privacy-Aware Market Basket Data Set Generation: A Feasible Approach for Inverse Frequent Set Mining

Abstract

Association rule mining has received a lot of attention in the data mining community and several algorithms were proposed to improve the performance of association rule or frequent itemset mining. The IBM Almaden synthetic data generator has been commonly used for performance evaluation. One recent work shows that the data generated is not good enough for benchmarking as it has very different characteristics from real-world data sets. Hence there is a great need to use real-world data sets as benchmarks. However, organizations hesitate to provide their data due to privacy concerns. Recent work on privacy preserving association rule mining addresses this issue by modifying real data sets to hide sensitive or private rules. However, modifying individual values in real data may impact on other, non-sensitive rules. In this paper, we propose a feasible solution to the NP-complete problem of inverse frequent set mining. Since solving this problem by linear programming techniques is very computationally prohibitive, we apply graph-theoretical results to divide the original itemsets into components that preserve maximum likelihood estimation. We then use iterative proportional fitting method to each component. The technique is experimentally evaluated with two real data sets and one synthetic data set. The results show that our approach is effective and efficient for reconstructing market basket data set from a given set of frequent itemsets while preserving sensitive information.