An increasing demand for customer-specific products is a major challenge for manufacturing companies. In many cases, companies attempt to satisfy this demand by increasing the number of product variants. In those companies, cost-oriented production processes have to be transformed into flexible workshop or island production structures in order to be able to produce this variety. This leads to an increasing complexity of production and subsequently planning. In order to reliably meet due dates, it is necessary to improve the quality of planning. This paper presents an approach for predicting transition times, the times between two production steps, by employing machine learning methods. In particular, the influence of the modelling of production knowledge of experienced employees on the prediction quality compared to a pure optimization of the methods' parameters is investigated.