Reinforcement Learning-Based Dynamic Medical Resource Allocation for Hospital Operating Cost Optimization
Efficient hospital resource allocation is essential for reducing operating costs under dynamic patient demand and changing clinical conditions. Existing methods, including Deep Neural Networks (DNN), Deep Q-Networks (DQN), and Policy Gradient Reinforcement Learning, often struggle to capture long-term temporal dependencies, limiting allocation efficiency. This study proposes Tx-D3QN, a hybrid framework that integrates a Double DQN with Transformer-based encoding to model both temporal and state-based hospital dynamics. The model is evaluated using the Kaggle Hospital Management Dataset (HMD), containing patient admissions, bed occupancy, staff allocation, treatment duration, and waiting times. After data preprocessing and feature optimization, Tx-D3QN achieved a 17.6% reduction in operational costs, 14.3% lower patient waiting times, and 96.1% resource utilization, outperforming DQN, DNN, and Policy Gradient Reinforcement Learning. These findings demonstrate the effectiveness of the proposed framework for intelligent and cost-efficient hospital resource management.
Authors
- Jiaqi Huang (ORCID: https://orcid.org/0009-0005-3576-9009)
- Qingjuan Liao
Institutions
- Nanning University (CN)
Publication Details
- Journal
- International Journal of Computational Intelligence and Applications
- Published
- 2026-11-12
- DOI
- https://doi.org/10.1142/s1469026826500677
- Primary Topic
- Healthcare Operations and Scheduling Optimization
- Type
- article
- Field-Weighted Citation Impact
- 0.00