This study compared nine machine-learning methods for creating surrogate models of an agent-based model (ABM) of social care provision in the UK. Neural networks outperformed other methods in accuracy, especially with larger datasets, but required longer training times. Gradient-boosted trees and non-linear SVM also demonstrated good performance with faster runtimes, suggesting machine learning can be a viable approach for efficient surrogate modeling of ABMs.

The authors acknowledge funding from various organizations, including UKRI EPSRC, UKRI Research England, and the Children's Liver Disease Foundation. However, they state that the funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. No other potential conflicts of interest were identified.

This study presents a robust evaluation of machine learning methods for surrogate modeling of ABMs, offering valuable insights for researchers working with complex simulations. The comparison across multiple methods and datasets, along with the detailed analysis of results, strengthens the paper's contribution. While some limitations exist regarding the scope of the study, the overall methodology and findings are valuable and warrant a strong rating.