Context prediction has been recognized as an enabler for proactive pervasive services that anticipate future situations already ahead of time. Traditional context predictors are limited by their agnostic view on the targeted application domain when analysing context histories of past user behaviour. Awareness about the processes in which an entity is involved can provide rich information to foresee future context changes more accurately. We present an approach for context prediction in pervasive environments that are characterized by context-aware workflows. In order to benefit from the explicit knowledge about human behaviour in these environments, we devise a context predictor that learns the relationship of context changes with the flow of activities performed by humans. This relationship is encoded as a probabilistic state transition system that can be explored to determine the most likely paths of future context occurrences. Our evaluation shows that our enhanced predictor is able to extract patterns from context histories that are inaccessible to history-only predictors and significantly improves the prediction accuracy.