Scholarlink

<p>This study presents a data-driven framework to identify and evaluate key variables contributing to four major distresses in Continuously Reinforced Concrete Pavement (CRCP): longitudinal cracking, spalling, punchouts, and transverse cracking. A Random Forest regression approach was applied to 395 observations extracted from the Long-Term Pavement Performance (LTPP) database, incorporating structural, climatic, and traffic-related inputs. Feature importance analysis revealed that temperature, Annual Average Daily Truck Traffic (AADTT), and Layer 3 thickness consistently ranked among the top predictors across multiple distress types. Spalling was highly predictable using only a few variables (R² &gt;0.85), whereas longitudinal and transverse cracking required more features to reach moderate accuracy, indicating complex, multifactorial mechanisms. Punchouts showed moderate predictability, influenced primarily by freeze index and heavy truck traffic. Sequential feature addition analysis confirmed that model performance improved significantly with the top-ranked variables, with diminishing returns thereafter. The proposed framework supports the development of parsimonious, interpretable models that can inform efficient CRCP design and maintenance strategies across varying environmental and operational conditions.</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

Random forest-based frame work for multi-distress prediction in CRCP: a feature importance approach

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月2日

查看全文
  • Ali Alnaqbi,
  • Ghazi G. Al-Khateeb,
  • Waleed Zeiada,
  • Muamer Abuzwidah

摘要

This study presents a data-driven framework to identify and evaluate key variables contributing to four major distresses in Continuously Reinforced Concrete Pavement (CRCP): longitudinal cracking, spalling, punchouts, and transverse cracking. A Random Forest regression approach was applied to 395 observations extracted from the Long-Term Pavement Performance (LTPP) database, incorporating structural, climatic, and traffic-related inputs. Feature importance analysis revealed that temperature, Annual Average Daily Truck Traffic (AADTT), and Layer 3 thickness consistently ranked among the top predictors across multiple distress types. Spalling was highly predictable using only a few variables (R² >0.85), whereas longitudinal and transverse cracking required more features to reach moderate accuracy, indicating complex, multifactorial mechanisms. Punchouts showed moderate predictability, influenced primarily by freeze index and heavy truck traffic. Sequential feature addition analysis confirmed that model performance improved significantly with the top-ranked variables, with diminishing returns thereafter. The proposed framework supports the development of parsimonious, interpretable models that can inform efficient CRCP design and maintenance strategies across varying environmental and operational conditions.