Phase-only spatial light modulators are ideal for the generation of beam splitter profiles to parallelize a variety of laser processes. A novel approach for the calculation of phase holograms is proposed to achieve a highly accurate power distribution over all spots. The Iterative Fourier Transform Algorithm (IFTA) is extended by the use of different machine learning methods, which are trained in an open camera-feedback loop. After the training phase, improvement of the beam splitting accuracy is then validated experimentally. The advantage of the presented approach is shown by comparing it to the standard IFTA algorithm. Finally, use of the approach is demonstrated through metal marking with an ultrashort pulse laser.