A hybrid reinforcement learning and knowledge graph framework for financial risk optimization in healthcare systems

Effective financial risk management in healthcare systems requires intelligent decision-making that balances treatment quality with cost efficiency. This paper proposes a novel hybrid framework that integrates reinforcement learning (RL) with knowledge graph-augmented neural networks to optimize billing decisions while preserving diagnostic accuracy. Patient profiles are encoded using a combination of structured features, deep latent representations, and semantic embeddings derived from a domain-specific knowledge graph. These enriched state vectors are used by an RL agent trained using Deep Q-Networks (DQN) or Proximal Policy Optimization (PPO) to recommend billing strategies that maximize long-term reward, reflecting both financial savings and clinical validity. Experimental results on real and synthetic healthcare datasets demonstrate that the proposed model outperforms traditional regressors, deep neural networks, and standalone RL agents across multiple evaluation metrics, including cost prediction error, diagnostic classification accuracy, cumulative reward, and average billing reduction. An ablation study confirms the complementary contributions of each architectural component. This work highlights the value of combining data-driven learning with structured medical knowledge to enable context-aware, cost-efficient decision-making in complex healthcare environments.