Investigation of attack strategies on geoprivacy with spatial obfuscation

Abstract

This research looks into the current state of spatial obfuscation algorithms and investigates attack strategies to circumvent them. A set of over 37.000 cyclist tracks are obfuscated using a variety of obfuscation algorithms, after which two categories of attack strategies were applied to reconstruct the original tracks based on the obfuscated tracks. These attack strategies were based on heuristic approaches on the one hand, with a deep learning approach to privacy attacks on the other. Using an evaluation measure determining the overlap between actual tracks and predicted tracks, each attack strategy was evaluated, revealing the applied deep learning approach not to be suitable as an attack strategy in its current form, and showing heuristic methods to function better. However, these methods are still unable to recover the original track completely, and further research is required to get attack strategies suitable to evaluate the performance of spatial obfuscation algorithms.