With the increased use of mobile devices, such as smartphones, users create a multitude of sensor data. These are used by so called location-based service (LBS), e.g. navigation or weather apps, to provide a service to the user. At the same time, such sensor data and in particula GPS information is privacy-sensitive as it gives away private attributes of the user, for example, their home or work address, religion or education. In this context, points of interest (POI) are of special interest, which refer to locations where a user spends a certain amount of time (such as home).

On the other hand, mobility data can provide interesting insights into mobility patterns which can help amongst others with urban planning, for example by analyzing traffic jams and planning new roads to relieve traffic. In order to being able to use this data for such applications but at the same time assuring the privacy of individuals, privacy-preserving mechanisms have been devised. These mechanisms aim to protect the privacy, for example by cloaking, disturbing or deleting data while minimizing the utility degradation of the data.

POIs seem to have a great impact on the private information that is revealed about a person. Therefore, in the bachelor thesis an approach is to be tested, that can identify POIs in a trajectory of a user in order to understand which parts of a trajectory need special attention with regards to protection. To this end an AutoEncoder is to be trained on trajectories that do not contain any such POIs so that „normal“ mobility behaviour is learned. This data can, for example, be generated with a routing service, such as openrouteservice. Via dimension reduction techniques and reconstruction error analysis it is to be analyzed whether POIs can be detected in real user trajectory data with this approach.