Control of flow separation over swept and dihedral wings, particularly at high angles of attack, is a key challenge in aerodynamic design. This study investigates the effect of simultaneous variations in wing sweep angle (0°–45°) and dihedral angle (0°–10°) on the performance of rectangular vortex generators (VGs) as a passive flow control method. A hybrid framework based on computational fluid dynamics (CFD) and machine learning was developed. Numerical simulations were performed using the k−ω SST turbulence model in a 2.5-dimensional domain with periodic boundary conditions, and an overset mesh approach was employed for parametric modeling of VGs. To reduce computational cost and efficiently explore the design space, a Gaussian Process Regression (GPR) surrogate model with an RBF-ARD kernel was trained, and single- and multi-objective optimization were conducted to determine the optimal VG position (x/c) and relative height (h/δ) for three selected wing configurations. The results show that optimal VG placement is strongly dependent on wing geometry, and a fixed VG configuration cannot be generalized for all cases. The highest improvement was achieved for the intermediate geometry (sweep angle of 29° and dihedral angle of 5°), where the optimal point (h/δ=1.18, x/c=0.12) resulted in a 27% reduction in drag coefficient and a 40% increase in lift-to-drag ratio compared with the clean wing. Independent validation showed prediction errors below 2% for lift coefficient and 2.6% for drag coefficient, confirming the reliability of the proposed framework. This study provides a data-driven design guideline for geometry-dependent optimization of vortex generators.