Machine Learning-Driven Tri-Level Hyper-Heuristic Selection With Adaptive Move Acceptance for Composing Medical Crew Scheduling Problem

Driven by the growing demand for adaptive and automated search control in combinatorial optimization, hyper-heuristic frameworks have become an important approach for handling heterogeneous problem instances. This work investigates the Composing Medical Crews with Equity and Efficiency (CMCEE) problem, which aims to maximize the minimum crew efficiency subject to a diversity constraint. In this study, a tri-level hyper-heuristic framework (TL-HH) was proposed to solve the CMCEE problem by coordinating decision-making across multiple search dimensions. First, a multi-stage selection structure is established to dynamically control the choice of the local search algorithm, neighborhood operator, and move acceptance rule within a unified optimization process. Second, a ranking-based regulation mechanism is introduced to evaluate the historical contribution of each low-level component and update the candidate pools accordingly. Third, a learning-driven acceptance strategy is designed to determine solution replacement decisions and improve the balance between diversification and intensification during a search. Fourth, an experimental protocol was developed to analyze the individual and combined effects of the three decision layers, providing insight into their coordinated interactions. Fifth, the convergence behavior and computational characteristics were examined to assess the stability and reliability of the proposed framework. Experiments were conducted on 80 benchmark instances of various sizes. The results demonstrate that TL-HH achieves consistent improvements in solution quality and robustness compared with the established hyper-heuristics and classical meta-heuristics for the CMCEE problem. These findings confirm the effectiveness of coordinated multi-level decision control for constrained optimization tasks and provide practical insights for adaptive search design in crew-composition problems.