A Pufferfish Privacy Mechanism for the Trajectory Clustering Task.

Monitoring people’s trajectories can help with many data mining tasks. For example, clustering analysis of users’ trajectories can infer where people work and where they live. However, as far as users are concerned, no one wants their privacy to be compromised so that third parties can benefit from it. The commonly used method of privacy protection today is differential privacy, but differential privacy does not have significant advantages when dealing with correlated data. Pufferfish privacy can be used to address the privacy protection of correlated data for this reason. Our work aims to protect the locations that are extracted from trajectories using clustering methods. To achieve this goal, we first use the DBSCAN algorithm to cluster the trajectories of users, mark the locations where users have stayed, and preserve the chronological order. Privacy requirements for this issue are then specified in the Pufferfish framework. The data is then processed using a mechanism that implements a privacy framework. Finally, the utility is evaluated through experiments.