On Leveraging Machine Learning in Sport Science in the Hypothetico-deductive Framework.

Area of Science:

• Sport Science
• Machine Learning
• Statistical Methods

Background:

• Supervised machine learning (ML) offers powerful predictive algorithms but often lacks transparency.
• Explainable ML and interpretable ML have emerged to address the "black box" nature of ML.
• The hypothetico-deductive framework is central to scientific inquiry, relying on hypothesis testing against data.

Purpose of the Study:

• To examine the fundamental differences between ML algorithms and statistical methods.
• To propose how supervised ML can augment, rather than replace, statistical methods in sport science.
• To provide guidance on the cautious integration of ML into the scientific workflow.

Main Methods:

• Comparative analysis of supervised ML and statistical methods.
• Examination of explainable and interpretable ML within the hypothetico-deductive framework.
• Case studies demonstrating the integration of supervised ML into exploratory analysis.

Main Results:

• Supervised ML algorithms and statistical models differ fundamentally in motivation and approach, despite addressing similar problems (y = f(x) + ε).
• While transparent ML methods increase understanding, they do not equate to statistical methods.
• Supervised ML can be valuable for exploratory analysis in sport science but requires cautious application.

Conclusions:

• Supervised machine learning should augment, not replace, statistical methods in sport science research.
• Integrating ML requires caution to leverage its strengths (e.g., complex pattern fitting) while avoiding pitfalls (e.g., misguided recommendations).
• Properly applied, supervised ML can enhance the hypothetico-deductive framework in sport science, but misuse mirrors statistical p-value hacking.