Hybrid Supervised–Reinforcement Learning Framework for Adaptive Decision Intelligence in Dynamic Environments

Abstract

Adaptive decision-making in dynamic and uncertain environments remains a significant challenge in artificial intelligence. Traditional supervised learning models excel in pattern recognition but lack adaptability, while reinforcement learning (RL) provides dynamic decision-making capabilities but often suffers from slow convergence and high sample complexity. This research proposes a hybrid supervised–reinforcement learning framework that integrates the strengths of both paradigms to enable efficient and adaptive decision intelligence. The proposed framework combines supervised learning for initial policy approximation with reinforcement learning for continuous optimization in changing environments. By leveraging labeled data for rapid model initialization and reward-driven learning for policy refinement, the hybrid approach improves learning efficiency, adaptability, and overall performance. Experimental evaluation demonstrates that the hybrid framework achieves faster convergence and higher accuracy compared to standalone supervised or reinforcement learning models. Additionally, the model exhibits strong robustness in non-stationary environments where data distributions evolve over time. The study also investigates optimization strategies such as experience replay, policy regularization, and adaptive reward shaping to enhance system performance. Results indicate that the hybrid approach effectively balances exploration and exploitation, making it suitable for applications in robotics, autonomous systems, finance, and intelligent decision support systems. This work contributes a scalable and flexible framework for next-generation adaptive AI systems capable of operating in complex real-world environments.