The sizing and shape optimization of truss structures is a challenging mixed discrete-continuous optimization problem due to nonlinear structural constraints and strong coupling between design variables and structural responses. In this paper, an improved Best–Mean–Worst–Random (IBMWR) algorithm is presented for the sizing and shape optimization of truss structures subject to stress and displacement constraints. This population-based metaheuristic balances the exploitation of high-performing solutions with controlled diversification through mean-, worst-, and random-guided search strategies. While the original Best–Mean–Worst–Random (BMWR) algorithm was applied to optimization tasks in metal casting processes [1], the IBMWR extends its application to structural optimization by incorporating a hybrid constraint-handling approach. This approach integrates an efficient constraint handling method [2], an adaptive penalty control strategy, and a structurally informed repair-based feasibility mechanism guided by critical stress and displacement responses. Sizing variables are selected from a discrete set of available cross-sectional areas, and shape variables are defined as continuous nodal coordinates within specified geometric limits. The objective of the optimization problem is to minimize the total structural weight. The performance of IBMWR is evaluated using the two well-known benchmark truss problems. Numerical results show that the proposed method yields results that are competitive with or superior to those reported for other optimization methods in the literature, including particle swarm optimization [3], differential evolution [4], the artificial bee colony algorithm [5], the medalist learning algorithm [6], and circulatory system-based optimization [7]. Furthermore, statistical performance indices, such as the best, mean, worst, and standard deviation of optimized weights from independent runs, demonstrate the robustness and reliability of the proposed algorithm for constrained structural optimization problems involving mixed discrete and continuous design variables.