Nanoporous metals, built out of complex ligament networks, can be produced with an additional level of hierarchy. The resulting complexity makes modeling of the mechanical behaviour computationally expensive and time consuming.  Surrogate models, predicting the mechanical behaviour of the lower level of hierarchy of FE beam models, are a promising approach to overcome these limitations. Therefore, as a first step, we compared data-driven and hybrid artificial neural networks, as well as data-driven support vector machines,  regarding their potential for the prediction of yield surfaces. All considered methods were well suited and resulted in relative errors <4.5%. However, support vector machines showed the highest generalization and accuracy in 6D stress space and outside the range of the used training data sets.