PPDAM

Abstract

Data mining is a process that analyzes voluminous digital data in order to discover hidden but useful patterns from digital data. However, the discovering of such hidden patterns has statistical meaning and may often disclose some sensitive information. As a result, privacy becomes one of the prime concerns in the data-mining research community. Since distributed association mining discovers association rules by combining local models from various distributed sites, breaching data privacy happens more often than it does in centralized environments. In this work, we present a methodology that generates association rules without revealing confidential inputs such as statistical properties of individual sites, and yet retains a high level of accuracy in the resultant rules. One of the important outcomes of the proposed technique is that it reduces the overall communication costs. Performance evaluation of our proposed method shows that it reduces the communication cost significantly when we compare it with other well-known, distributed association-rule-mining algorithms. Nevertheless, the global rule model generated by the proposed method is based on the exact global support of each item set and hence diminishes inconsistency, which indeed occurs when global models are generated from partial support count of an item set.