Pavement Performance Prediction using Machine Learning: Supervised Learning with Tree-Based Algorithms
Tiago Tamagusko, Adelino Ferreira
Abstract
This study employs supervised machine learning tree-based algorithms to predict the performance of flexible pavements, in particular the International Roughness Index (IRI). Three algorithms, namely Decision Tree, Random Forest (RF), and eXtreme Gradient Boosting (XGBoost) were applied to assess pavement quality. The data to develop the machine learning models were sourced from the Long-Term Pavement Performance InfoPave database, selecting 55 experimental sections of asphalt concrete on granular and asphalt concrete on bound bases. The study considered only pavements without maintenance or rehabilitation. Likewise, models were trained using the structural number, annual average daily truck traffic, precipitation and temperature, with IRI as the target variable. The results demonstrated that the XGBoost model outperformed the others, achieving an R-squared value of 0.98, while the Random Forest model achieved an R-squared value of 0.95. These findings indicate the potential of tree-based algorithms in predicting pavement performance, offering valuable insights for future research.