Scholarlink

<p>In many countries including Palestine, the Pavement Condition Index (PCI) is determined by an exhausted and time-consuming process that involves thorough visual inspections and are costly to obtain. This paper develops and evaluates the effectiveness of predicting the PCI by proposing a machine learning framework that predicts PCI classes rather than scores by systematically excluding distresses as inputs, offering a more efficient and practical approach to PCI prediction. A Machine Leaning approach is proposed to model the complex relationship between the PCI and other measured variables including age (year), pavement layer thickness (m), ROW (m), ADT (veh/day), heavy duty percentage (%), number of lanes, and distress features. Features including Alligator Cracking, Rut Depth, and Longitudinal Cracking were found to have highest impact on pavement condition assessment. These distresses are known to be direct indicators of structural deterioration and overall road performance, making their high importance scores expected. From practical standpoint, the results show that both Random Forest and XGBoost are well-suited for pavement condition monitoring, especially in identifying well-maintained and critically deteriorated roads with an overall accuracy of 82% for both. Transportation agencies can use these models to automate pavement assessments, prioritize maintenance tasks effectively, and lower the costs associated with manual inspections.</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

Predicting Pavement Condition Index (PCI) Using Machine Learning: A Distress Exclusion Approach for Practical Decision-Making

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Amjad Issa,
  • Huthaifa I. Ashqar

摘要

In many countries including Palestine, the Pavement Condition Index (PCI) is determined by an exhausted and time-consuming process that involves thorough visual inspections and are costly to obtain. This paper develops and evaluates the effectiveness of predicting the PCI by proposing a machine learning framework that predicts PCI classes rather than scores by systematically excluding distresses as inputs, offering a more efficient and practical approach to PCI prediction. A Machine Leaning approach is proposed to model the complex relationship between the PCI and other measured variables including age (year), pavement layer thickness (m), ROW (m), ADT (veh/day), heavy duty percentage (%), number of lanes, and distress features. Features including Alligator Cracking, Rut Depth, and Longitudinal Cracking were found to have highest impact on pavement condition assessment. These distresses are known to be direct indicators of structural deterioration and overall road performance, making their high importance scores expected. From practical standpoint, the results show that both Random Forest and XGBoost are well-suited for pavement condition monitoring, especially in identifying well-maintained and critically deteriorated roads with an overall accuracy of 82% for both. Transportation agencies can use these models to automate pavement assessments, prioritize maintenance tasks effectively, and lower the costs associated with manual inspections.