Modeling of Rutting and Wheel Path Longitudinal Cracking Performance Interaction Using a Heterogeneous Ensemble Machine-Learning Approach

This study aimed to develop heterogeneous ensemble machine-learning models to numerically evaluate the distress interaction mechanism of rutting and wheel-path longitudinal cracking (WPLC) in asphalt pavements. To achieve this, a total of 206 pavement sections were selected from the long-term pavement performance (LTPP) database, and the particle swarm optimization algorithm was utilized to optimize the developed rutting–cracking base models including the random forest, support vector regression, backpropagation neural network, and extreme gradient boosting models. New machine-learning models were constructed using a heterogeneous ensemble method to combine the two base models with better prediction performance. The corresponding results yielded a relatively high accuracy, with coefficient of determination (R2) values of 0.8868 and 0.8465 for the rutting and WPLC models, respectively. The sensitivity analyses revealed that the properties of the asphalt surface layer had a more significant impact on both rutting and WPLC performance than other pavement structure layers. More importantly, further analyses demonstrated that increased rut depth was associated with more severe WPLC in the field, indicating an interrelationship between these two pavement distress types in the pavement wheel path.