Dynamic Patient Scheduling in Hospitals Using Variable Length Non-Dominated Sorting Genetic Algorithm III

Effective patient scheduling in hospitals is crucial for optimizing resource use and improving patient care. Traditional methods often struggle to balance patient preferences, hospital constraints, and varying patient loads. This study explores the III genetic algorithm without dominant sorting with variable length (VL-NSGA III) for dynamic patient scheduling and compares it with Particle Swarm Optimization (PSO), Multi-Objective Particle Swarm Optimization (MOPSO), Objective Decomposition Particle Swarm Optimization (ODPSO), and Genetic Algorithm without Dominant Sorting II with Best Fitness Evaluation (Nsga2bfe). The problem formulation considers dynamic patient arrivals and hospital constraints, requiring flexible solutions. VL-NSGA III generates high-quality nondominant solutions tailored to dynamic scheduling scenarios. The evaluation used simulator-based scoring over a 36-day period, with synthetic patient data simulating real hospital conditions. The simulation modeled a hospital with multiple departments, specializations, and rooms, considering factors such as room capacity, patient arrival rates, and service duration. Evaluation metrics included set coverage (C-metric) to assess dominance among solution sets, hypervolume (HV) to measure objective space coverage, and convergence to measure proximity to the true Pareto front. The study ran multiple simulation scenarios with varying patient arrival rates, service durations, and hospital capacities to test the algorithm's robustness and adaptability. The results showed that VLNSGA III excelled at generating non-dominated solutions with superior set coverage, achieving a value of 1 against PSO, MOPSO, ODPSO, and Nsga2bfe, indicating complete dominance. ODPSO achieved the highest hypervolume, closely followed by MOPSO and PSO. Notably, MOPSO demonstrated partial dominance over PSO with 0.7 coverage and over ODPSO with 0.8333. ODPSO showed partial dominance over PSO and MOPSO with coverage values of 0.6333 and 0.7333, respectively. Nsga2bfe exhibited partial dominance over VL-NSGA III with a coverage value of 0.03333 while fully dominating PSO and MOPSO. The dominant set coverage of VL-NSGA III highlighted its robustness and adaptability in dynamic patient scheduling scenarios, despite lower hypervolume values compared to ODPSO, MOPSO, and PSO. This underscores the importance of considering both set coverage and hypervolume metrics when evaluating algorithm performance for complex scheduling problems.