Predictive Quality of Service in Vehicular Communication Systems Models, Datasets, and Open Challenges
TL;DR
This survey provides a structured review of QoS prediction approaches spanning machine learning, deep learning, reinforcement learning, protocol-level evaluation, and security-trust mechanisms, and outlines promising research directions for developing adaptive, explainable, and privacy-preserving QoS prediction models for emerging 5G/6G V2X systems.
Abstract
Reliable Quality of Service (QoS) prediction is fundamental to the performance of next-generation Vehicle-to-Everything (V2X) communication, particularly in 5G-enabled cloud and edge environments that support autonomous driving, cooperative safety, and time-critical mobility services. This survey provides a structured review of QoS prediction approaches spanning machine learning, deep learning, reinforcement learning, protocol-level evaluation, and security-trust mechanisms. Existing methods are analyzed with respect to their modeling strategies, required datasets, spatio-temporal dependencies, and scalability under high-mobility conditions. A comprehensive dataset survey synthesizes real-world, open-access, and simulation-based resources, highlighting key variables such as SINR, RSRP/RSRQ, mobility patterns, traffic density, delay, and throughput that influence predictive accuracy. The review identifies persistent challenges, including limited multi-modal datasets, poor cross-scenario generalization, computational overhead, weak interpretability, and the absence of unified multi-metric prediction frameworks. Based on these insights, the paper outlines promising research directions for developing adaptive, explainable, and privacy-preserving QoS prediction models for emerging 5G/6G V2X systems.