Machine Learning-Driven Patient Scheduling in Healthcare: A Fairness-Centric Approach for Optimized Resource Allocation

Effectively managing queues and scheduling re-sources to minimize wait times is challenging. For example, in healthcare facilities, managing waiting times for multiple types of patients while optimally utilizing resources without causing patient dissatisfaction requires intervention by experienced staff. In the future, online consultations and treatment options are likely to increase. This will lead to a requirement for automation of the allocation of resources and the possible use of AI-based optimization techniques. In this work, we look at real-world patient arrival patterns and multiple allocation techniques to understand the impact of optimized allocation on the fairness of the experience for patients. We also look at the possibility of tuning the hyperparameters of machine learning techniques to balance optimized allocation while also considering fairness. In this work, we use Reinforcement Learning (RL) and Deep Learning RL, along with a Modified Queuing System using CloudSim. We used data from a healthcare facility with variations to model task arrival (patient arrival) and, depending on severity, needing to be handled by edge nodes and cloud-based servers (two types of doctors). These techniques were compared based on some important parameters. The results indicate that our Modified Queuing System ensures complete fairness and balances waiting times between categories of patients but does not attempt to optimize resource utilization. RL can reduce wait times but may not ensure fairness among patients based on the time of arrival. To address this fairness issue, the Deep Learning RL approach was applied, allowing for tunable fairness in patient allocation.