Scholarlink

<p>This study explores machine learning (ML) applications in rigid pavement performance evaluation from 2010 to 2025, following PRISMA guidelines. Through comprehensive analysis of 268 studies across major databases, this research evaluates the predictive capabilities of various ML algorithms, identifies critical input parameters, and synthesizes methodological trends in pavement condition assessment. The study reveals that artificial neural networks and support vector machines are most frequently applied, while ensemble methods achieve superior predictive performance with the highest R<sup>2</sup> values. Traffic-related inputs dominate ML applications, followed by environmental factors and structural parameters. The Pavement Condition Index and International Roughness Index emerge as the most predicted performance indicators. Key findings indicate significant research gaps including limited long-term validation, insufficient model interpretability focus, and lack of standardized evaluation frameworks. A novel Rigid Pavement Performance Machine Learning framework incorporating a Weighted Performance Index is proposed to address these limitations. The analysis confirms ensemble methods' superiority in balancing predictive accuracy, interpretability, and data requirements, providing evidence-based recommendations for practitioners and identifying critical future research directions in pavement engineering applications.</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

Machine Learning Approaches in Rigid Pavement Performance Evaluation

  • 期刊论文

  • 发表日期：

    2026年4月9日

查看全文
  • Jitendra Singh Yadav,
  • Kamal Singh

摘要

This study explores machine learning (ML) applications in rigid pavement performance evaluation from 2010 to 2025, following PRISMA guidelines. Through comprehensive analysis of 268 studies across major databases, this research evaluates the predictive capabilities of various ML algorithms, identifies critical input parameters, and synthesizes methodological trends in pavement condition assessment. The study reveals that artificial neural networks and support vector machines are most frequently applied, while ensemble methods achieve superior predictive performance with the highest R2 values. Traffic-related inputs dominate ML applications, followed by environmental factors and structural parameters. The Pavement Condition Index and International Roughness Index emerge as the most predicted performance indicators. Key findings indicate significant research gaps including limited long-term validation, insufficient model interpretability focus, and lack of standardized evaluation frameworks. A novel Rigid Pavement Performance Machine Learning framework incorporating a Weighted Performance Index is proposed to address these limitations. The analysis confirms ensemble methods' superiority in balancing predictive accuracy, interpretability, and data requirements, providing evidence-based recommendations for practitioners and identifying critical future research directions in pavement engineering applications.