Synthetic Data for Sharing and Exploration in High-Performance Sport: Considerations for Application

Area of Science:

• Sports and Exercise Science
• Data Science
• Computational Statistics

Background:

• Synthetic data offer privacy-preserving alternatives for research and practice.
• Applications are emerging in clinical and medical fields, supporting open science.
• Mathematical procedures generate synthetic data to address specific research challenges.

Purpose of the Study:

• To evaluate the synthpop package for generating synthetic athlete monitoring data.
• To provide an educational primer on synthetic data generation in sports and exercise science.
• To discuss the applicability of sequential tree-based algorithms for sports data.

Main Methods:

• Utilized the R synthpop package on a professional football dataset.
• Applied seven simulation conditions with varying model constraints.
• Assessed global and specific utility metrics for each simulation.

Main Results:

• All simulations achieved high global utility, indicating overall dataset similarity.
• Simpler simulation conditions (1 and 2) showed higher specific utility.
• More complex simulations resulted in lower specific utility compared to simpler ones.

Conclusions:

• Three conceptual models exist for synthetic data generation: analysis, generation, and true process models.
• Misalignments between these models can introduce bias and compromise utility.
• Researchers must document synthetic data generation purpose, models, predictors, and limitations for sports science applications.