Predictive QoS in V2X Communication: Enhancing Throughput Prediction with Machine Learning in Urban Environments

The article Improving QoS Prediction in Urban V2X Networks by Leveraging Data from Leading Vehicles and Historical Trends, published on April 23, 2025, by Sanket Partani, Michael Zentarra, Anthony Kiggundu, and Hans D. Schotten, explores how integrating data from lead vehicles enhances the precision of QoS predictions in urban V2X networks using machine learning techniques.

The study investigates Predictive Quality of Service (PQoS) using machine learning models to predict downlink throughput in urban Vehicle-to-Everything (V2X) communication systems. Using the Berlin V2X cellular dataset, features from both ego and lead vehicles were analyzed to improve throughput predictions. Results demonstrate enhanced model performance when incorporating lead vehicle data, with improvements observed across various machine learning algorithms, indicating a model-agnostic benefit.

In the evolving landscape of autonomous driving and road safety, vehicle-to-everything (V2X) communication is a critical component. Ensuring high quality of service (QoS) is paramount, as any lapse could lead to accidents or system failures. Researchers are employing machine learning models such as XGBoost, CNNs, and LSTMs to predict QoS metrics like throughput, thereby enhancing reliability.

The study utilized a dataset sourced from multiple vehicles in Berlin, incorporating spatial, temporal, and network features. Spatial features included location data, while temporal aspects encompassed time of day and day of the week. Network features involved signal strength and latency. These elements were used to train machine learning models, with performance evaluated using metrics such as Mean Absolute Error (MAE), Symmetric Mean Absolute Percentage Error (SMAPE), and Root Mean Square Error (RMSE).

The choice of models was strategic: XGBoost for its efficiency with structured data, CNNs for potential pattern recognition in sequential data, and LSTMs for handling temporal dependencies. Interestingly, XGBoost outperformed the others with lower error rates, suggesting that the dataset may lack strong long-term dependencies, making gradient boosting more effective.

Future research directions include generating synthetic data to supplement real-world measurements, particularly for rare scenarios, and fine-tuning models for specific times or areas. This approach could enhance adaptability to varying conditions such as rush hour or urban vs. rural environments. Practical challenges include privacy concerns with real-time data sharing and model adaptability in dynamic environments. Additionally, the integration of these models into existing V2X systems presents considerations regarding whether they should run on edge devices for low-latency decisions or in the cloud.

It is crucial to address concept drift, where patterns change due to new vehicles or traffic conditions. The need for efficient model integration into current infrastructure underscores the importance of robust implementation strategies.

This research highlights the effectiveness of machine learning in enhancing QoS within V2X communication, with a focus on practical applications. The findings underscore the significance of model selection and data adaptability in advancing autonomous driving technologies, offering valuable insights for future research and application development.