A Privacy-Preserving Trajectory Mining Model

In this paper, we consider user privacy issues in location-based social networks (LBSNs). To encourage users to upload personal trajectory data for the benefit of society, we propose a novel privacy-preserving model to protect user trajectory data during the transmission and in the data center (the cloud). The proposed encryption model consists of a Gaussian-whiten encoder and a rotation matrix for the encryption key. The advantage of this model is that the encrypted data can be used for machine learning tasks and generative tasks, even it is not decrypted. We design a feasible interaction strategy to share the model between users and the cloud, and protect the encryption keys from cloud and external eavesdroppers. We apply our model to the trajectory dataset and feed the encrypted data to two classification tasks. Simulation results show that the proposed privacy-preserving model can protect user privacy at an acceptable cost of accuracy loss for classification tasks. Meanwhile, it outperforms classical differential privacy (DP) methods in terms of classifier accuracy loss and privacy preservation ability.