Core Concepts
This paper presents a novel approach to accurately predict Quality of Service (QoS) metrics, such as packet delivery ratio (PDR) and throughput, in NR-based Vehicle-to-Everything (V2X) scenarios using various machine learning (ML) methods with a nested cross-validation scheme.
Abstract
The paper focuses on predicting QoS in NR-V2X communications using machine learning (ML) techniques. It utilizes a dataset generated through simulations using SUMO and 5G-Lena, which includes parameters such as modulation and coding scheme (MCS), distance-to-base station, signal-to-interference-plus-noise ratio (SINR), and user datagram protocol (UDP) packet size.
The key highlights and insights are:
The paper employs a novel nested cross-validation approach, which prevents information leakage from parameter selection into hyperparameter selection, resulting in more robust and reliable ML models compared to traditional cross-validation.
Several ML methods are evaluated, including support vector regression (SVR), artificial neural network (ANN), gradient boosting regression (GBR), random forest (RF), light gradient boosting method (LGBM), and categorical boosting regression (CBR).
The performance of the ML models is assessed using metrics such as mean absolute error (MAE), root mean square error (RMSE), and R-squared (R2) score.
The results show that ensemble learning methods, particularly RF, GBR, and CBR, outperform ANN and SVR in predicting throughput and PDR, demonstrating the effectiveness of these approaches in the context of NR-V2X QoS prediction.
The nested cross-validation approach ensures the robustness of the ML models, making them suitable for real-time applications in the evolving field of wireless communication.
Stats
The dataset used in this study was generated through simulations using SUMO and 5G-Lena, and it includes the following key parameters:
Modulation and coding scheme (MCS)
Distance-to-base station
Signal-to-interference-plus-noise ratio (SINR)
User datagram protocol (UDP) packet size