IIS-IPS: A privacy-preserving framework for distance-based mining

University Of Maryland Baltimore County, Baltimore MD

Investigators

Abstract

The goal of this project is to develop a novel framework for preserving privacy in distance based data mining. This research helps protect the privacy of sensitive data such as medical exam reports and, at the same time, allow the discovery of interesting patterns using distance based mining algorithms. The approach in this project consists of a pre-processing step to de-correlate the data and an additive perturbation step to provide worst-case privacy guarantees. This approach also provides the necessary and sufficient conditions for such guarantees. In addition, modifications are made to existing distance-based data mining algorithms so that these algorithms can run accurately on the perturbed data. The results of this project provide privacy preserving data mining techniques with both worst-case privacy guarantees and high accuracy of mining results. These techniques have potential applications in many areas, especially in critical areas such as law enforcement where finger prints, foot prints, and facial images are matched using distance-based algorithms. This research is also linked to educational goals through dissemination of the results to K-12 educational and outreach programs, undergraduate and graduate courses, and interdisciplinary conferences and workshops. The results of this project will be disseminated via the project website: http://www.is.umbc.edu/privacy_research