Product and process quality is playing an increasingly important role in the competitive success of manufacturing companies. To ensure a high quality level of the produced parts, the appropriate selection of parameters in manufacturing processes plays in important role. Traditional approaches for parameter setting rely on rule-based schemes, expertise and domain knowledge of highly skilled workers or trial and error. Automated and real-time adjustment of critical process parameters, based on the individual properties of a part and its previous production conditions, have the potential to reduce scrap and increase the quality. Different machine learning methods can be applied for generating parameter estimation models based on experimental data. In this paper, we present a comparison of linear and symbolic regression methods for an adaptive parameter setting approach. Based on comprehensive real-world data, collected in a long-term study, multiple models are generated, evaluated and compared with regard to their applicability in the studied approach for parameter setting in manufacturing processes.