The transition toward climate-neutral mobility necessitates highly efficient manufacturing processes for electric traction drives. A critical step in the production of high-power density stators is the laser welding of rectangular copper wires, a process characterized by complex interdependencies between numerous parameters. This paper presents a systematic multi-objective Bayesian optimization approach to identify optimal welding parameters with minimal experimental effort. The methodology utilizes a surrogate model based on Gaussian Processes to efficiently explore a seven-dimensional parameter space, including feed rate, laser power distribution, and complex weld geometries. An initial training dataset was generated using Latin Hypercube Sampling to ensure a broad and uniform distribution across the parameter space. The optimization targets three conflicting objectives: minimizing process time and electrical contact resistance, while maximizing mechanical peel strength. The results demonstrate that a weighted objective function strategy effectively identifies non-dominated solutions on the Pareto front. Validation through metallographic cross-sections and Partial Dependence Plots confirms a significant improvement in connection quality compared to initial trials, achieving a reduction in electrical resistance and a substantial increase in mechanical strength. This research underlines the potential of data-driven machine learning methods to significantly enhance stability and resource efficiency in high-precision laser welding applications for electric motor manufacturing.