Industrial Automation (IA) applications require deterministic communication channels to ensure a reliable operation. As the wireless medium is a shared medium used by many other wireless technologies, a deterministic medium access method (MAM) is necessary. An improvement of the coexistence behavior can be achieved by applying adaptive MAMs, but they cannot meet any real-time demands. A promising approach to meet real-time as well as coexistence demands are cognitive MAMs. We evaluate the performance of three different cognitive MAMs which differ in the probabilistic prediction model: Two methods are based on Markov modelling (MM) and one method is based on an auto-regressive (AR) model. The MAMs are experimentally evaluated in a worst case wireless measurement scenario.