There has been a rise in the number of successful machine learning models serving as a key to identify and utilize linkages along the process-structure-property-performance chain for greatly different challenges in the domains of materials mechanics throughout the last decades. Consideration of physical laws in data-driven modelling has recently been proven to offer improved prediction performances and higher generalization while using less data, in comparison to using either physics-based or data-driven modelling techniques alone. A simulation-assisted machine learning framework is explained in this contribution using the example of friction surfacing, a solid-state layer deposition method that may be utilized for additive manufacturing as well as for repair or coating applications. The goal is to employ machine learning approaches to predict and analyse the impact of process parameters and ambient factors, such as substrate and backing material properties, on process behaviour and deposit geometry. A particular focus is placed on the effects of maximum temperature on prediction objectives during the process supplied by a numerical heat transfer model. Several machine learning algorithms are reviewed in order to take advantage of their varied capabilities and choose the optimal one for the given problem and data. Moreover, the targets' feature dependencies are evaluated based on game-theory related Shapley Additive Explanation values. The experimental data sets were generated with as few samples as possible via two distinct designs of experiments, one for the variation of process parameters and the other for the variation of substrate and backing material properties. The goal was to also represent the cross parameter space between the two separate spaces, which was accomplished and allowed for an approximate 45% reduction in the number of trials when compared to carrying out an experiment design that includes both spaces.