Improving scalability of privacy preserving algorithms

Data mining techniques reveal patterns in large databases that may be strategically relevant. Organizations prepare the data before participating in data sharing agreements in order to avoid revealing tactically important insights to external organizations. Viable techniques that preserve the privacy of strategically significant frequent item sets are essential in the protection of a competitive advantage. This research improves the scalability of a frequent item set hiding algorithm through a partitioning heuristic that decomposes an exact hiding algorithm problem formulation into multiple smaller sections that are processed separately to generate partial extensions of the database. The smaller data extensions are then combined to form one extension that reduces the statistical significance of all sensitive frequent item sets without changing the statistical significance of the remaining data. By requiring less processing resources, the partitioning heuristic improves the scalability of the exact hiding algorithm.