A hybrid machine learning method of support vector regression with particle swarm optimization for predicting IRI in continuously reinforced concrete pavement

Abstract For efficient maintenance planning and long-term performance assessment of continuously reinforced concrete pavement (CRCP), an accurate estimate of the International Roughness Index (IRI) is necessary. In order to forecast IRI using data taken from the Long-Term Pavement Performance (LTPP) database, this study proposes a hybrid machine learning model that combines Support Vector Regression (SVR) and Particle Swarm Optimization (PSO). Incorporating structural, climatic, and traffic-related variables, 395 observations from 33 CRCP sections were used. The PSO algorithm was employed to optimize SVR hyperparameters, resulting in enhanced model accuracy and stability. The proposed PSO-SVR model achieved outstanding predictive performance with an average RMSE of 0.04116 and an R 2 of 0.99058 across fivefold cross-validation, outperforming benchmark models including Decision Tree, Random Forest, and XGBoost. By highlighting important input characteristics affecting IRI, variable importance analysis and 3D interaction plots improved the interpretability of the model even more. The outcomes show the PSO-SVR framework's superiority and dependability, underscoring its potential as a strong decision-support tool for pavement management and performance forecasting in rigid pavement systems.