In injection molding the determination of appropriate process parameters to achieve a desired part quality can be a time-consuming and challenging task. The simulation offers the possibility to map the injection molding process and to make predictions about the process behavior. However, there is a gap between simulation and physical experiment due to the complexity of the injection molding process, the dynamic machine reaction, the limitation of simulation techniques, their simplification and approximation error. In order to reduce the test effort and to increase the predictability of the produced part quality, machine learning models can be used. To generate data for these models, a test plan needs to be created, which then can be simulated and directly run on the machine. Based on a similarity analysis the first machine learning model can be trained to predict, which test points can be substituted by simulation and which must be run on the machine. A second, optimized model is then trained to predict the part quality also considering the process parameters. The results of the research will be presented at the PPS conference.