Dummy Trajectory Generation Scheme Based on Deep Learning

Nowadays, the traditional dummy trajectory generation algorithm used to protect users’ trajectory privacy usually uses statistical methods to build trajectory model. Because the human mobility model is a complex equation, it is difficult to use mathematical methods to model, so the established trajectory model can not consider the human mobility model which restricts the formation of trajectory. Therefore, traditional dummy trajectory generation algorithms can not defend against data mining attacks based on in-depth learning. In this paper, LSTM (Long Short-Term Memory) is used to design the real and dummy trajectory discriminator. Experiments show that data mining based on deep learning can eliminate more than 95% of the algorithm generated trajectories, and the error rate of real trajectory is less than 10%. We restrict the traditional dummy trajectory generation algorithm to human mobility model, and design a dummy trajectory generation strategy so that the generated trajectory can defend against multiple recognition attacks.