<p>Optimizing truss structures is a key challenge in structural engineering, requiring methods that balance design quality, computational efficiency, and solution reliability. This paper proposes the adaptive surrogate-based hybrid optimizer (ASBHO), a novel framework that integrates adaptive surrogate modeling with hybrid optimization to effectively address the aforementioned challenges. ASBHO leverages a bagged decision tree and a neural network as adaptive surrogates, guided by a feasibility detection phase using an improved constrained lower confidence bound criterion with leave-one-out cross-validation (CLCB-LOOCV). This phase efficiently identifies feasible regions with minimal computational effort, enabling the optimization process to focus on both global exploration and local refinement. A dynamic switching mechanism between two optimizers—social learning particle swarm optimization (SLPSO) and particle swarm optimization (PSO)—further enhances search efficiency, reducing computational cost without compromising accuracy. The proposed method is tested on six distinct truss structures, with two of them additionally evaluated under different conditions, resulting in a comprehensive assessment across eight benchmark problems. Compared with state-of-the-art algorithms, ASBHO demonstrates superior performance, reducing the average structural weight by 1% while decreasing computational time by an impressive 76%. These results underscore ASBHO’s potential as a powerful and efficient tool for real-world structural optimization.</p>