An adaptive metamodeling framework for simulation-based optimization: a healthcare planning case study

Simulation-based optimization has become an important decision-support approach for improving the performance of complex stochastic systems. However, conventional optimization methods often require a large number of simulation experiments, resulting in high computational costs and limiting their practical applicability. This study proposes an innovative Adaptive Metamodeling-based Simulation Optimization (AMSO) framework that efficiently combines Bagging-based Gradient Boosted Trees, hyperparameter optimization, and metaheuristics to accelerate the search for high-quality solutions while reducing computational effort. The proposed framework is validated through a real-world case study involving the planning of a temporary hospital in Brazil, modeled using Discrete Event Simulation. The AMSO framework demonstrated significant advantages over traditional methods. Specifically, AMSO outperforms the Efficient Global Optimization (EGO) approach and achieves similar outcomes to the Genetic Algorithm (GA) while requiring substantially fewer simulation experiments and lower computational time. In the case study, the optimized solutions improved resource planning and reduced patient Length of Stay. Beyond the healthcare application, the proposed framework provides an adaptive, efficient, and robust simulation-based optimization approach that can be applied to a wide range of complex systems requiring computationally expensive simulation optimization.