Due to complex network conditions, autonomous vehicles possibly do not have time to react appropriately when experiencing unexpected quality of service (QoS) degradation, which can impact the safety of road participants. Deep learning is considered to be an efficient way to provide accurate predictions about QoS based on vehicle and network information for vehicle-to-vehicle (V2X) communication applications and notify applications to adapt to upcoming QoS changes by the network proactively. In this paper, we examine a multivariate multi-step multi-output prediction method based on Long- and Short-term Time-series network (LSTNet) to predict QoS Key Performance Indicators (KPIs) as end-to-end latency and Packet Reception Ratio. Joint simulations based on NS-3 and SUMO demonstrate our approach’s effectiveness in QoS prediction.