Identifying Latent State-Transition Processes for Individualized Reinforcement Learning