Supporting Streaming Data Anonymization with Expressions of User Privacy Preferences

Abstract

Mining crime reports in real-time is useful in improving the response time of law enforcement authorities in addressing crime. However, limitations on computational processing power and in-house mining expertise make this challenging, particularly so for law enforcement agencies in technology constrained environments. Outsourcing crime data mining offers a cost-effective alternative strategy. Yet outsourcing crime data raises the issue of user privacy. Therefore encouraging user participation in crime reporting schemes is conditional on providing strong guarantees of personal data protection. Cryptographic approaches make for time consuming query result generation, so the preferred approach is to anonymize the data. Mining real-time crime data as opposed to static data facilitates fast intervention. To achieve this goal, Sakpere and Kayem presented a preliminary solution based on the notion of buffering. Buffering improves on information loss significantly in comparison with previous solutions. In this paper, we extend the Sakpere and Kayem result to support user privacy expressions. We achieve this by integrating a three-tiered user-defined privacy preference model in data stream process. The three-tiered model offers a simple and generic approach to classifying the data without impacting negatively on information loss. Results from our proof-of-concept implementation indicate that incorporating user privacy preferences reduces the rate of information loss due to misclassification.