Abstract: To address the suboptimal process parameter optimization in laser-arc hybrid welding (LAHW) of marine thin steel plates under small-sample conditions, an enhanced data-driven multi-objective optimization method for LAHW process parameters is proposed. Necessary experimental data were acquired based on orthogonal tests, with welding speed, laser-arc distance, laser power, and wire feed speed serving as optimization variables, while maximum welding residual stress and deformation angle were designated as optimization objectives. To establish accurate mapping from process parameters to optimization objectives, a multi-objective sub-stacking strategy-based multi-model ensemble (MOSSMME) prediction method was introduced, and generative adversarial networks (GANs) were employed to augment the original data to enhance the generalization performance of the prediction model. On this basis, the second-generation non-dominated sorting genetic algorithm-Ⅱ (NSGA-Ⅱ) was adopted to solve this multi-objective optimization problem. Finally, relevant LAHW experiments were conducted for validation. The results demonstrate that the proposed prediction method combining MOSSMME and GANs effectively improves prediction accuracy and robustness under small-sample conditions; the average error between optimization results and experimental measurements is merely 3.72%, confirming the feasibility of the proposed enhanced data-driven multi-objective optimization method, which holds significant reference value for the optimization of welding process parameters.