Feature-Based Interpretable Reinforcement Learning based on State-Transition Models