Scholarlink

<p>Predicting the international roughness index (IRI) with accuracy is crucial for long-term maintenance planning, efficient pavement management, and guaranteeing ride quality and safety. Using a hybrid machine learning framework, the current research tries to produce a predictive model that is both reliable and comprehensible for estimating IRI in continuously reinforced concrete pavement (CRCP). By tuning important hyperparameters including learning rate, tree depth, and the number of boosting iterations, the suggested model combines gradient boosting machines (GBM) with particle swarm optimization (PSO) to improve predictive accuracy. A dataset of 395 observations from 33 CRCP sections obtained from the long-term pavement performance (LTPP) database was used for the research. Twenty input features covering structural, climatic, traffic, and initial performance characteristics were used to predict measured IRI values. The model’s performance was evaluated using five-fold cross-validation and compared against benchmark models including unoptimized GBM, linear regression, support vector regression (SVR), random forest, and artificial neural networks (ANN). The PSO-GBM model achieved superior results with a mean RMSE of 0.0398 and R<sup>2</sup> of 0.9910, demonstrating excellent prediction accuracy. Feature importance and sensitivity analyses identified initial IRI, layer thicknesses, pavement age, and truck traffic as the most influential variables. Partial dependence plots (PDPs) were used to improve interpretability and visualize variable effects on IRI predictions. These results confirm that the PSO-GBM model is a robust and practical tool for data-driven pavement evaluation and has strong potential for integration into pavement management systems.</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 PSO-tuned gradient boosting model for accurate IRI prediction in continuously reinforced concrete pavements

  • 期刊论文

  • 发表日期：

    2025年9月8日

查看全文
  • Ali Alnaqbi,
  • Waleed Zeiada,
  • Ghazi Al-Khateeb

摘要

Predicting the international roughness index (IRI) with accuracy is crucial for long-term maintenance planning, efficient pavement management, and guaranteeing ride quality and safety. Using a hybrid machine learning framework, the current research tries to produce a predictive model that is both reliable and comprehensible for estimating IRI in continuously reinforced concrete pavement (CRCP). By tuning important hyperparameters including learning rate, tree depth, and the number of boosting iterations, the suggested model combines gradient boosting machines (GBM) with particle swarm optimization (PSO) to improve predictive accuracy. A dataset of 395 observations from 33 CRCP sections obtained from the long-term pavement performance (LTPP) database was used for the research. Twenty input features covering structural, climatic, traffic, and initial performance characteristics were used to predict measured IRI values. The model’s performance was evaluated using five-fold cross-validation and compared against benchmark models including unoptimized GBM, linear regression, support vector regression (SVR), random forest, and artificial neural networks (ANN). The PSO-GBM model achieved superior results with a mean RMSE of 0.0398 and R2 of 0.9910, demonstrating excellent prediction accuracy. Feature importance and sensitivity analyses identified initial IRI, layer thicknesses, pavement age, and truck traffic as the most influential variables. Partial dependence plots (PDPs) were used to improve interpretability and visualize variable effects on IRI predictions. These results confirm that the PSO-GBM model is a robust and practical tool for data-driven pavement evaluation and has strong potential for integration into pavement management systems.