Optimizing punchout prediction in rigid pavement using a hybrid GA-SVR approach

A critical structural distress in Continuously Reinforced Concrete Pavement (CRCP), punchouts frequently result in expensive repairs and a shorter service life. For proactive maintenance, optimal rehabilitation planning, and long-term pavement performance, accurate punchout prediction models need to be developed. In this work, a hybrid machine learning method utilizing Support Vector Regression (SVR) optimized by Genetic Algorithm (GA) is presented for punchout occurrence prediction in CRCP. In order to incorporate structural, environmental, traffic, and performance variables, a dataset consisting of 395 observations from 33 pavement sections was taken from the Long-Term Pavement Performance (LTPP) database. Several models, including Linear Regression, Decision Tree, Random Forest, XGBoost, and Artificial Neural Network (ANN), were created and assessed using 5-fold cross-validation following preprocessing and normalizing process. With the highest R2 (0.96267) and the lowest mean RMSE (1.7139), the GA-SVR model exceeded all benchmark models, demonstrating its better predictive performance. Feature importance analysis identified pavement age, climate zone, and layer characteristics as dominant predictors, while sensitivity analysis highlighted the influence of precipitation, humidity, and traffic loading. Additionally, hyperparameter sensitivity evaluations demonstrated that kernel scale and epsilon had significant effects on SVR performance, validating the importance of GA-based optimization. Three-dimensional surface plots were used to visually interpret the interaction of key variables, enhancing model transparency. Overall, the proposed GA-SVR model offers a reliable and interpretable tool for forecasting punchouts in CRCP, supporting more informed maintenance planning and pavement management strategies.