We present a nondestructive and data-driven method for predicting eggshell thickness using impact-based sensing and machine learning. A custom low-speed impact module was developed to simulate mechanical responses of duck eggshells, and the Hertzian contact theory was employed to interpret deformation behavior. Three machine learning models—random forest (RF), XGBoost, and K-nearest neighbors (KNN)—were implemented and optimized with metaheuristic algorithms, including particle swarm optimization (PSO). Among them, the RF model obtained by PSO demonstrated superior prediction accuracy with an R2 of 0.65155 and a mean squared error (MSE) of 0.00044. The proposed approach offers a scalable, cost-effective alternative to traditional eggshell assessment techniques and can be readily integrated into industrial egg grading systems to enhance food quality monitoring and reduce product waste.