Forecasting asphalt concrete's dynamic modulus with a hybrid machine learning approaches
Liu Wei,Lu Xinrong,Chen Liang,Deng Daping,Mehdi Kouhdarag,Liang Tongxiang
Abstract
The study aimed to forecast asphalt concrete's dynamic modulus (|E∗|, |G∗|) using hybrid machine learning, combining MLP (Multi-Layer Perceptron), SVM (Support Vector Machine), DT (Decision Tree), RF (Random Forest), LR (Logistic Regression), and AdaBoost classifiers to understand dataset complexities. With 2,238 data points from 2010 to 2023 and diverse asphalt concrete samples, a 70-30 train-testing (70% for training and 30% for testing) split ensured thorough analysis. This dataset's diversity was further enriched by incorporating asphalt concrete samples with varying geometries of 100 mmx200 mm and 100 mmx150 mm (diameterxheight), contributing to a more holistic understanding of the material's behavior. Evaluation metrics included confusion matrix, MAE, RMSE, and R2. Results highlighted predictive models' impact on |E∗| and |G∗| values, especially MLP's accuracy (0.964) and precision (0.955), making it reliable for engineers and researchers. MLP also excelled in the testing dataset (accuracy: 0.964, precision: 0.903). SVM followed with 0.880 accuracy and 0.852 precision. These outcomes reinforced the MLP model's reliability and underscored its potential as an asset in predicting |E∗| and |G∗| modulus values, affirming its practical applicability in geotechnical studies and research endeavors.
Liu Wei — 1) Jiangxi University of Science & Technology, Ganzhou 341000, Jiangxi, China, 2) Gannan University of Science & Technology, Ganzhou 341000, Jiangxi, China
Lu Xinrong, Chen Liang, Deng Daping — Gannan University of Science & Technology, Ganzhou 341000, Jiangxi, China
Mehdi Kouhdarag — Civil Engineering Department, Malekan Branch, Islamic Azad University, Malekan 5561788389, Iran
Liang Tongxiang — Jiangxi University of Science & Technology, Ganzhou 341000, Jiangxi, China
PDF Viewer
Preview is limited to the first 3 pages. Sign in to access the full PDF.
