Location obfuscation and distance-based attacks on private trajectories: an experimental evaluation on real trajectory data sets

Official URL: http://risc01.sabanciuniv.edu/record=b1606651 (Table of Contents)

With the wide availability of GPS-enabled mobile devices, spatio-temporal data is being collected and stored for providing location-based services or for data analytics. Location-based advertisement, market research, and data mining are just some of the motivations for collecting spatio-temporal data. However, location data is also very sensitive since it may reveal information about the data subject such as his/her political views, religion, state of health, and various personal preferences, which is considered private information. One of the approaches to protect sensitive location data is obfuscation. In this thesis, we have implemented two location obfuscation techniques, performed an analytical and experimental study to investigate how e ective they are on a state of the art attack algorithm designed for spatio-temporal data. In the attack scenario, given a set of known trajectories, and a distance matrix composed of known pairwise distances between trajectories, adversary tries to approximate the target trajectory and then extract information about absence or presence of the trajectory in a given area. We used obfuscation techniques to hide information around predefined sensitive places such as hospitals, medical centers. We then used obfuscated data on the attack. Experimental results show that the applied obfuscation methods do not help protecting the privacy of users in sensitive areas in case of spatio-temporal trajectories that follow a regular pattern. We observed that the attack method works successfully because the obfuscation techniques do not scatter the sensitive points far enough from sensitive places and the linearity of the trajectory is preserved.