Enhancing Q-Value Updates in Deep Q-Learning via Successor-State Prediction