Scholarlink

<p>The resilient modulus is a crucial parameter in the design and analysis of flexible pavement systems, as it directly influences the pavement’s structural response to traffic loading. Accurately predicting the resilient modulus of fine-grained soils is challenging due to the complex, nonlinear, and stress-dependent behavior of these materials under cyclic loading conditions. This study proposes a novel voting-based ensemble approach that integrates Gradient Boosting (GB) and Multilayer Perceptron (MLP) neural networks to predict the resilient modulus of fine-grained subgrade soils. The paper compiled a comprehensive database of resilient modulus test results and associated soil properties, drawing from both laboratory and field-based measurements. Data from 3907 soil samples were gathered from the Long-Term Pavement Performance (LTPP) website. The proposed hybrid approach combines the strengths of gradient boosting and multilayer perceptron models to capture the intricate relationships between soil properties, environmental factors, and resilient modulus. This voting-based ensemble technique aims to provide more accurate and reliable predictions compared to individual machine learning models, whose performance can be heavily dependent on the specific dataset and conditions. Additionally, Principal Component Analysis (PCA) and Shapley Value (SV) analysis are employed to provide insights into the model’s performance and the relative importance of input features. The finding suggests that the ensemble approach was successful, achieving a 97.8% R² score, and that silt content, axial stress, and clay content have a significant impact on predicting resilient modulus values. The proposed voting-based ensemble approach demonstrated substantial improvements in resilient modulus prediction, with a 44.62% reduction in Mean Absolute Error (MAE) and a 42.74% reduction in Root Mean Squared Error (RMSE) compared to the stand-alone multilayer perceptron model. The findings from this research are expected to contribute to a better understanding of the resilient response of fine-grained subgrade soils, leading to more accurate pavement design, improved infrastructure performance, and enhanced sustainability in transportation 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

Resilient Modulus Prediction for Fine-Grained Soils Using a Novel Voting-Based Ensemble Approach: Integrating Gradient Boosting and Multilayer Perceptron Neural Networks

  • 期刊论文

  • 发表日期：

    2025年12月18日

查看全文
  • Mohammad Ali Khasawneh,
  • Hiren Mewada,
  • Ahmad Ali Khasawneh,
  • Mohammed Nayeemuddin,
  • Samer Rababah

摘要

The resilient modulus is a crucial parameter in the design and analysis of flexible pavement systems, as it directly influences the pavement’s structural response to traffic loading. Accurately predicting the resilient modulus of fine-grained soils is challenging due to the complex, nonlinear, and stress-dependent behavior of these materials under cyclic loading conditions. This study proposes a novel voting-based ensemble approach that integrates Gradient Boosting (GB) and Multilayer Perceptron (MLP) neural networks to predict the resilient modulus of fine-grained subgrade soils. The paper compiled a comprehensive database of resilient modulus test results and associated soil properties, drawing from both laboratory and field-based measurements. Data from 3907 soil samples were gathered from the Long-Term Pavement Performance (LTPP) website. The proposed hybrid approach combines the strengths of gradient boosting and multilayer perceptron models to capture the intricate relationships between soil properties, environmental factors, and resilient modulus. This voting-based ensemble technique aims to provide more accurate and reliable predictions compared to individual machine learning models, whose performance can be heavily dependent on the specific dataset and conditions. Additionally, Principal Component Analysis (PCA) and Shapley Value (SV) analysis are employed to provide insights into the model’s performance and the relative importance of input features. The finding suggests that the ensemble approach was successful, achieving a 97.8% R² score, and that silt content, axial stress, and clay content have a significant impact on predicting resilient modulus values. The proposed voting-based ensemble approach demonstrated substantial improvements in resilient modulus prediction, with a 44.62% reduction in Mean Absolute Error (MAE) and a 42.74% reduction in Root Mean Squared Error (RMSE) compared to the stand-alone multilayer perceptron model. The findings from this research are expected to contribute to a better understanding of the resilient response of fine-grained subgrade soils, leading to more accurate pavement design, improved infrastructure performance, and enhanced sustainability in transportation systems.