Scholarlink

<p>Real-time roadside unit (RSU) tracking in vehicle-to-vehicle (V2V) networks is challenged by high vehicular density (100–25,000 nodes), dynamic mobility patterns, and noisy communication signals. Conventional methods lack scalability, nonlinear feature modelling, and uncertainty quantification, leading to unreliable performance under varying traffic conditions. To address these issues, we propose a hybrid probabilistic ensemble framework integrating Bayesian Deep Neural Network (BDNN)-based feature transformation, Extreme Gradient Boosting (XGBoost) classification, and a Gaussian Discriminant Filter (GDF) for posterior refinement. The BDNN performs adaptive feature extraction while modelling epistemic uncertainty. The transformed features are processed by XGBoost to capture nonlinear decision boundaries efficiently. Subsequently, the GDF applies class-conditional Gaussian modelling to refine posterior probabilities, improving robustness against noise and distributional shifts. Evaluated on real-time simulated vehicular mobility datasets across varying traffic densities, the proposed model achieves over 95% classification accuracy with stable scalability and improved precision–recall balance. Comparative results demonstrate superior accuracy, uncertainty handling, and computational efficiency over conventional and standalone ensemble approaches. The framework provides a scalable and uncertainty-aware RSU tracking solution for large-scale intelligent transportation systems (ITS).</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A hybrid machine learning approach for scalable and uncertainty-aware RSU-based tracking and performance optimization in VANETs

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月22日

查看全文
  • Krishna Komaram,
  • Nagarjuna Karyemsetty

摘要

Real-time roadside unit (RSU) tracking in vehicle-to-vehicle (V2V) networks is challenged by high vehicular density (100–25,000 nodes), dynamic mobility patterns, and noisy communication signals. Conventional methods lack scalability, nonlinear feature modelling, and uncertainty quantification, leading to unreliable performance under varying traffic conditions. To address these issues, we propose a hybrid probabilistic ensemble framework integrating Bayesian Deep Neural Network (BDNN)-based feature transformation, Extreme Gradient Boosting (XGBoost) classification, and a Gaussian Discriminant Filter (GDF) for posterior refinement. The BDNN performs adaptive feature extraction while modelling epistemic uncertainty. The transformed features are processed by XGBoost to capture nonlinear decision boundaries efficiently. Subsequently, the GDF applies class-conditional Gaussian modelling to refine posterior probabilities, improving robustness against noise and distributional shifts. Evaluated on real-time simulated vehicular mobility datasets across varying traffic densities, the proposed model achieves over 95% classification accuracy with stable scalability and improved precision–recall balance. Comparative results demonstrate superior accuracy, uncertainty handling, and computational efficiency over conventional and standalone ensemble approaches. The framework provides a scalable and uncertainty-aware RSU tracking solution for large-scale intelligent transportation systems (ITS).