Basalt Fiber Asphalt Concrete (BFAC) is an environmentally friendly and durable material with potential road, bridge, and infrastructure construction applications. This study investigates the application of Machine Learning (ML) models, specifically the classical Gradient Boosting (CGB) algorithm, in conjunction with metaheuristic algorithms, to predict the Marshall Stability (MS) and optimize the design of BFAC mixtures. The model is trained and tested on a comprehensive dataset of experimental samples, taking into account various input parameters, including basalt fiber (BF) properties, asphalt binder characteristics, and aggregate gradation. Hyperparameter tuning is employed to enhance the model's predictive performance using metaheuristic algorithms such as Particle Swarm Optimization (PSO), Hunger Games Search (HGS), and Bald Eagle Search (BES) and compared regarding the convergence and computational efficiency. The findings demonstrate that BES outperforms other algorithms, achieving the highest performance. The CGB-BES model is then applied to three optimization scenarios, focusing on maximizing the MS while minimizing BF and asphalt binder content. Post-processing and interpretation of the results reveal the importance of combining ML and materials engineering expertise. By highlighting the synergy between CGB-BES model and domain-specific knowledge, materials engineers can effectively optimize the mixtures and improve the design and performance of BFAC.