Differentially Private Anomaly Detection with a Case Study on Epidemic Outbreak Detection

Anomaly detection is an important problem that has been studied in a variety of application domains, ranging from syndrome surveillance for epidemic outbreaks to intrusion detection in computer networks. The data collected from individual users contain sensitive information, such as health records and network usage data, and thus need to be transformed prior to the release for privacy preservation. In this paper, we propose a novel framework for anomaly detection with differential privacy. Real-time private user data can be aggregated and perturbed to guarantee privacy, while the posterior estimate is released continuously for anomaly detection tasks. Our framework is not limited to any specific application domains. We illustrate the sensitivity analysis and evaluate our framework in the context of syndrome surveillance. Empirical results with simulated data sets confirm the effectiveness of our solution while providing provable privacy guarantee.