Scholarlink

<p>Accurate path loss prediction is critical for reliable Vehicle-to-Vehicle (V2V) communication in highway environments, where traditional empirical models fail to capture traffic-dependent propagation dynamics. This paper presents a two-stage machine learning framework that explicitly integrates traffic density classification into path loss prediction at 5.9 GHz—addressing a common limitation of existing approaches that treat varying traffic conditions uniformly. Our hierarchical architecture first classifies traffic density (low/high) using basic propagation measurements, achieving 97.8% accuracy despite 13:1 class imbalance, then performs density-conditioned regression with engineered features capturing distance-density interactions. Systematic evaluation on 26,043 highway measurements compares six ML algorithms against five empirical models, with Kalman filtering preprocessing demonstrating 3.5–5.6<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\times \)</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> superior noise reduction. Rigorous ablation studies confirm that explicit density modeling reduces prediction error by 10.5%, capturing 75.6% of oracle performance. Our best model (random forest, RMSE = 2.388 dB, <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(R^2\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> = 0.964) achieves 42–57% error reduction over empirical approaches in mixed-density scenarios while maintaining real-time feasibility (54,993 predictions/second). These results provide concrete implementation guidelines for next-generation Cellular V2X and 6G vehicular networks requiring traffic-adaptive propagation modeling.</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 two-stage machine learning approach to path loss prediction at 5.9 GHz: benchmarking empirical and data-driven models for V2V highway scenarios

  • 期刊论文

  • 发表日期：

    2026年4月27日

查看全文
  • Hayfa Fhima,
  • Hanene Zormati,
  • Jalel Chebil,
  • Jamel Belhadj Taher

摘要

Accurate path loss prediction is critical for reliable Vehicle-to-Vehicle (V2V) communication in highway environments, where traditional empirical models fail to capture traffic-dependent propagation dynamics. This paper presents a two-stage machine learning framework that explicitly integrates traffic density classification into path loss prediction at 5.9 GHz—addressing a common limitation of existing approaches that treat varying traffic conditions uniformly. Our hierarchical architecture first classifies traffic density (low/high) using basic propagation measurements, achieving 97.8% accuracy despite 13:1 class imbalance, then performs density-conditioned regression with engineered features capturing distance-density interactions. Systematic evaluation on 26,043 highway measurements compares six ML algorithms against five empirical models, with Kalman filtering preprocessing demonstrating 3.5–5.6 \(\times \) × superior noise reduction. Rigorous ablation studies confirm that explicit density modeling reduces prediction error by 10.5%, capturing 75.6% of oracle performance. Our best model (random forest, RMSE = 2.388 dB, \(R^2\) R 2 = 0.964) achieves 42–57% error reduction over empirical approaches in mixed-density scenarios while maintaining real-time feasibility (54,993 predictions/second). These results provide concrete implementation guidelines for next-generation Cellular V2X and 6G vehicular networks requiring traffic-adaptive propagation modeling.