Research On Membership Inference On No-Time Aggregate Trajectory

With the popularity of mobile smart phones,location-based application services are more and more widely used,and application service parties collect a large amount of user movement trajectory data,which brings the risk of location privacy leakage to users.In order to better protect personal privacy,data publishers usually protect user data before data is released,and tend to publish strong-privacy-protection data without user identification,without time information and after differential privacy disturbance,which named No-Time Aggregate Trajectory.Performing member inference on no-time aggregate mobile trajectory data to determine the existence of target users,and thus discovering privacy leaks,has practical significance for further improving privacy protection methods.A scheme(Membership Inference on No-Time Aggregate Trajectory,MINTAT)is designed for the membership inference on no-time aggregation trajectory,for no-time information scenarios,only location information is used for member inference.MINTAT uses the known movement trajectory data of the past time period as the historical background knowledge data set,extracts the grid position features of the historical background knowledge data set,including the number,distance,and density,and uses the random forest model for machine learning to obtain good trained model.MINTAT can perform member inference on the target no-time aggregate trajectory in the future time period.During the inference process,it extracts the position distribution feature information consistent with the background knowledge data set,enters the previously trained machine learning model,and performs member inference prediction,obtain member-inference results.A large number of member inference experiments on no-time aggregate trajectory have been carried out.The experimental results show that the experimental results of MINTAT are good on sparse data sets and dense data sets;the results of MINTAT are better than the related work in most cases;MINTAT has a good adaptability to the mesh diameter parameter,privacy budget parameter,and background knowledge scale parameter.