17th International Fiber and Polymer Research Symposium

Home / 15th International Fiber and Polymer Research Symposium

Prediction of Marshall Strength in Polymer-Modified Thin-Layer Asphalt Using Decision Tree Regression and Bayesian Optimization

Authors :

Uğur ÖZVEREN¹ Ozan Derya KAYA² Ayserin Pelin YILMAZ³ Burce PEKER⁴ Idil Sena BAYRAK⁵ Kabil Bilgic⁶

1- Marmara University, Department of Engineering Faculty 2- Marmara University, Department of Engineering Faculty 3- Marmara University, Department of Engineering Faculty 4- Marmara University, Department of Engineering Faculty 5- Marmara University, Department of Engineering Faculty 6- Marmara University, Department of Engineering Faculty

Keywords :

Polymer-Modified Asphalt،Marshall Strength Prediction،Machine Learning،Decision Tree Regression،Hyperparameter Optimization

Abstract :

This study presents a machine learning approach for predicting the Marshall Strength of polymer-modified thin-layer asphalt mixtures using Decision Tree regression with Bayesian optimization. The research utilized a dataset of 45 samples with varying aggregate crushing strengths (80-100) and polymer percentages (5-15%), designed according to EN13108-2 standards. The model achieved exceptional accuracy with R² values of 0.9985 and 0.9214 for training and test sets, respectively. Feature importance analysis revealed that polymer percentage has a dominant influence (57.66%) on Marshall Strength compared to aggregate crushing strength (42.34%). The comprehensive evaluation using multiple regression metrics demonstrated the model's strong predictive capabilities, despite some evidence of overfitting indicated by the differences between training and test metrics (Training RMSE: 6.9572, Test RMSE: 48.2378). This approach provides a practical tool for optimizing asphalt mixture designs, potentially reducing the need for extensive laboratory testing while maintaining prediction accuracy. The methodology offers significant implications for the development of more durable and cost-effective pavement solutions, particularly in regions with varying climate conditions and heavy traffic loads.