In this paper, we propose the prediction system which uses the matchable situation decomposition technology in order to achieve a concept acquisition model which predicts various events in the real world. Our proposed prediction system extracts very regular partial situations which correspond to the concepts, constructs neural networks based on each concept, and predicts outputs based on the concepts selected by the integration unit. By simulations of using the card classification problem, we show that our proposed prediction system can predict from fewer events than other prediction systems.