Wireless traffic prediction has drawn increasing research interest because it can provide network optimization guidance. With the predicted information, one can preassign resources on demand and adaptively perform network congestion control. The efficiency of the utilized network is therefore enhanced. However, conducting wireless traffic prediction in the context of mobile scenarios, such as the Internet of Vehicles (IoV), is still challenging. The mobile nature of vehicles, which dynamically changes the topology of the constructed network, makes prediction difficult. This chapter focuses on implementing deep learning-based wireless traffic prediction in the IoV scenario. Section 2.1 proposes a novel method for matching the movement and communication behaviors of vehicles by merging two independent datasets containing the trajectories of vehicles and communication traffic volumes. Then, a novel STeP-UNet is proposed in Sect. 2.2, in which a spatiotemporal partial (STeP) convolutional neural network module is embedded to capture the cross-domain features of the observed wireless traffic pattern, and the UNet structure is utilized to realize skip connections from the front layer to the back layer to fuse different resolutions. The experimental results confirm the promising performance of the proposed model in Sect. 2.3, where a 4~8% performance improvement can be achieved over other benchmark methods.