Towards a simplification of models using regression trees.

Area of Science:

• Ecology
• Computational Biology
• Systems Biology

Background:

• Over-parametrization in scientific models complicates the identification of parameter-behavior relationships.
• Complex models often require numerous parameters, hindering interpretability and robustness.

Purpose of the Study:

• To present an empirical method for simplifying complex models by reducing the number of parameters.
• To provide an objective tool for modelers to decrease parameter ranges and identify parameter relationships.

Main Methods:

• Utilized regression trees to classify model outputs based on input parameters.
• Applied an empirical approach to systematically decrease model complexity.
• Demonstrated the method on a dynamic energy budget model of a mesopelagic bacterial ecosystem.

Main Results:

• Successfully simplified a complex ecological model, reducing its parameter set.
• The method facilitated objective parameter reduction and identification of parameter interdependencies.
• Achieved more robust parameter estimations, less complex model formulations, and fewer modeling assumptions.

Conclusions:

• The proposed empirical method offers a mathematical basis for model simplification.
• Benefits include improved parameter estimation, reduced complexity, and fewer assumptions.
• Challenges were identified, and potential solutions were discussed for broader applicability.