On the adversarial inconsistencies in crowdsourcing navigation systems

Crowdsourcing navigation systems (CNS) are now a vital component of intelligent transport systems (ITS). CNS allows crowd access to the navigational data. However, this openness renders it vulnerable to adversarial inconsistencies and attacks. Consequently, these issues can disrupt the routing decisions of ITS and thus decline the performance of data aggregation systems. Besides, the adversarial inconsistencies could mislead ITS and traffic control systems. Hence, there is a need for identifying and removing adversarial inconsistencies from CNS. Therefore, we propose a deep generative adversarial network (GAN) based model to classify and remove the adversarial inconsistencies, thereby enhancing the performance of CNS. The proposed model maps the CNS GPS trajectories to a multi-variate time series (MTS) model using a self-trained generator-discriminator framework for adversarial learning. Then, neural networks are used to classify adversarial GPS trajectories. We empirically verify the proposed GAN-based model through experiments. The results demonstrate the capability of the proposed model in identifying adversarial trajectories and improving the reliability of the CNS system.