Scholarlink

<p>Accurate estimation of optimum asphalt content is essential for achieving durable and high-performance hot mix asphalt (HMA) pavements. This study develops a hybrid machine learning–genetic algorithm (ML–GA) framework to predict the optimum asphalt content using a comprehensive dataset comprising aggregate gradations across multiple sieve sizes, kinematic viscosity, maximum specific gravity, bulk specific gravity, and Voids in Mineral Aggregate (VMA) values. To make the model more robust, data augmentation is used with the Gaussian Noise method, and its performance is tested via K-fold cross validation (CV). Four machine learning algorithms Random Forest (RF), Huber Regressor, Gradient Boosting (GBM), and Light Gradient Boosting Machine (LightGBM) were trained and evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R²) metrics. Based on CV results, the GBM model achieved the highest performance, and, yielding a test R² of 0.838, the model was subsequently optimized using a GA, which further improved performance. This process resulted in a test R² of 0.865, with reduced error metrics. Model interpretability using Shapley Additive Explanations (SHAP) revealed that fine aggregate fraction (No_200_passing) and coarse aggregate gradation (No_4_passing) were the most influential parameters on asphalt content. The proposed ML–GA hybrid model provides a reliable, interpretable, and data-driven alternative to conventional mix design methods in asphalt pavement engineering.</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

AI-Based Prediction of Optimum Binder Content in Asphalt Mixtures Using Machine Learning–Genetic Algorithm Hybrid Modeling

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Fatih Ergezer

摘要

Accurate estimation of optimum asphalt content is essential for achieving durable and high-performance hot mix asphalt (HMA) pavements. This study develops a hybrid machine learning–genetic algorithm (ML–GA) framework to predict the optimum asphalt content using a comprehensive dataset comprising aggregate gradations across multiple sieve sizes, kinematic viscosity, maximum specific gravity, bulk specific gravity, and Voids in Mineral Aggregate (VMA) values. To make the model more robust, data augmentation is used with the Gaussian Noise method, and its performance is tested via K-fold cross validation (CV). Four machine learning algorithms Random Forest (RF), Huber Regressor, Gradient Boosting (GBM), and Light Gradient Boosting Machine (LightGBM) were trained and evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R²) metrics. Based on CV results, the GBM model achieved the highest performance, and, yielding a test R² of 0.838, the model was subsequently optimized using a GA, which further improved performance. This process resulted in a test R² of 0.865, with reduced error metrics. Model interpretability using Shapley Additive Explanations (SHAP) revealed that fine aggregate fraction (No_200_passing) and coarse aggregate gradation (No_4_passing) were the most influential parameters on asphalt content. The proposed ML–GA hybrid model provides a reliable, interpretable, and data-driven alternative to conventional mix design methods in asphalt pavement engineering.