An Explicit Model for Soil Resilient Modulus Incorporating Freezing–Thawing Cycles Through Offspring Selection Genetic Algorithm (OSGA)

Determining soil resilient modulus for pavement design traditionally involves resource-intensive repeated load triaxial testing, prompting the need for a reliable and efficient prediction model. While previous research studies have explored evolutionary algorithms and genetic programming to create closed-form models, a significant gap exists in incorporating freezing–thawing cycles, a vital yet neglected factor affecting resilient modulus. This study addresses this gap by leveraging a comprehensive dataset of 1111 data points and introducing a novel explicit closed-form prediction equation. Utilizing the offspring selection genetic algorithm (OSGA), the proposed model not only demonstrates high accuracy, with a coefficient of determination (R²) = 0.93 and mean absolute error (MAE) of 6.54 MPa for the training set and R² = 0.92 and MAE = 6.91 MPa for the testing set, but also surpasses traditional black-box machine learning models by its interpretability, transparently, and the possibly of using it in routine designs. Additionally, sensitivity analysis revealed that freezing–thawing cycle count is the most influential parameter, followed by the weighted plasticity index and water content.