Context awareness and prediction are important for pervasive computing systems. The recently developed theory of context spaces addresses problems related to sensor data uncertainty and high-level situation reasoning. This paper proposes and discusses componentized context prediction algorithms and thus extends the context spaces theory. This paper focuses on two questions: how to plug-in appropriate context prediction techniques, including Markov chains, Bayesian reasoning and sequence predictors, to the context spaces theory and how to estimate the efficiency of those techniques. The paper also proposes and presents a testbed for testing a variety of context prediction methods. The results and ongoing implementation are also discussed.