An Improved Scheme for Privacy-Preserving Collaborative Anomaly Detection

The ubiquity of mobile sensing devices in the Internet of Things (IoT) enables an emerging data crowdsourcing paradigm called participatory sensing, where multiple individuals collect data and use a cloud service to analyse the union of the collected data. An example of such collaborative analysis is collaborative anomaly detection. Given the possibility that the cloud service is honest but curious, a major challenge is how to protect the participants' privacy. The scheme called Random Multiparty Perturbation (RMP) addresses this challenge by allowing each participant to perturb his/her tabular data by passing the data through a nonlinear function, and projecting the data to a lower dimension using a participant-specific random matrix. Here, we propose an improvement to RMP by introducing a new nonlinear function. The improved scheme is assessed in terms of its recovery resistance to the maximum a priori (MAP) estimation attack. Experimental results and preliminary theoretical analysis indicate that RMP is resistant to collusion attacks and has better recovery resistance to MAP estimation attacks compared to the original scheme. It also achieves a good trade-off between accuracy and privacy.