Predicting rutting in asphalt pavement of RIOH track using multiple machine learning approaches

This study develops predictive models for rutting progression using Multiple Linear Regression (MLR), Support Vector Regression (SVR), and Transformer architectures, based on data collected from the RIOHtrack testing facility. A three-stage data cleaning process is applied to enhance data quality, and principal component analysis is employed to reduce nine environmental variables to three key common factors. Rutting resistance indices (RIA and RIB), derived from calculations based on different pavement structures and materials, are used as model inputs in combination with environmental and traffic load variables. Of the 19 pavement structures, 16 are used for training and 3 for validation. The MLR model is excluded due to its limited capacity for capturing nonlinear relationships. The SVR model, optimized via Particle Swarm Optimization (PSO), demonstrates improved performance. Validation results reveal that the Transformer model surpasses the PSO-SVR by approximately 25%, achieving higher predictive accuracy (R² = 0.82 vs. 0.66) and generalization capability.