A deep learning model for secure cyber-physical transportation systems

In cyber-physical transportation systems (CPTS), moving vehicles and road sensors transmit traffic information to controllers via wireless networks. Because of the open property of the wireless communications, the CPTS are vulnerable to security attacks, e.g., malicious eavesdropping and jamming. Furthermore, for the smart eavesdropper and jammer, which can adjust their attack strategies according to the transmission power of information, it is a challenge to learn the strategy of a security attack, and further to adjust the transmission power against the attack. In this paper, we propose a deep learning model, Deep-SCPT, to learn such a kind of strategy (it can be considered as the feature of an attack). This model uses unsupervised learning to achieve an active learning process. Extensive experiments are carried out with 10 different datasets, and the results illustrate that the deep learning model achieves an average gain of 6% accuracy compared to state-of-the-art machine learning algorithms.