<p>With the variety of materials, printing processes and suppliers in additive manufacturing, the need to determine optimized process settings for the combinations in a resource-saving manner is increasing. Traditionally, experimental designs are set up to define the variable combinations to be tested. Mathematical models describing the interactions between input and output variables are created using methods of a design of experiment (DoE). Such modeled relations and predicted results can also be achieved using machine learning (ML) methods. In this paper, a detailed comparison between the two methods using an example in the field of additive manufacturing is accomplished, using the identical input data. Furthermore, quadratic or interaction effects of the input variables are considered and the results from the DoE are included to improve the modeling of the ML methods. In this study, the results of a response surface model as a representative of classical statistical DoE are compared with several ML methods. Fused layer modeling (FLM) is selected as the basis for this comparison with the tensile strength to be optimized. The results show that the variables Walline count, infill density, layer height, and flow are significant for the tensile strength. Quadratic effects and interactions between the variables are observed. When comparing the root mean squared error, the response surface model performs at least similar or even better than the Machine Learning (ML) algorithms. This is explained by the tested variables value combinations, which were selected in favor of the statistical design of experiments. A sequential combination of statistical design of experiments and machine learning methods improves the predictive properties of the machine learning models, while keeping the information transparency of the DoE and keeping the testing volume to a necessary minimum. </p>