Unlike paved roads, granular or unpaved roads are prone to rutting when subjected to overweight traffic such as Implements of Husbandry (IoH) or Superheavy Loads (SHL), and regular maintenance is essential to maintain the proper shape of the road cross-section. While most granular roads are designed for low-volume traffic, they often experience transportation of heavy farm products to the marketplace, leading to significant rutting damage and associated maintenance costs. There is therefore a growing demand for mechanistic analysis of such IoHs and SHLs, also known as superloads, to prevent severe rutting dam- age. To address this issue, this study developed Artificial Neural Network (ANN)-based surrogate models to quantify rutting damages, primary structural defects in granular roads that result in permanent deformation both of the top granular layer and the subgrade. The models considered various road structural parameters and commonly-encountered superload types used in the Midwestern region of the U.S. Optimized ANN models for granular roads under superloads were derived by comparing prediction accuracies of ANN models developed using varying numbers of hidden layers and neurons and applying different back- propagation algorithms such as Levenberg–Marquardt, Bayesian Regularization, BFGS Quasi-Newton, Resilient Backpropagation, and Scaled Conjugate Gradient. The prediction models developed through this study were shown to be superior in predicting rutting damages, suggesting their potential to provide a basis for the Mechanistic-Empirical design of granular roads and their integration into granular surface defect-prediction models in the future, forming a comprehensive prediction framework for analyzing granular road system.