Predicting special forces dropout via explainable machine learning.

Area of Science:

• Sports Science
• Performance Psychology
• Machine Learning Applications

Background:

• Effective selection is crucial for organizational success in sports, business, and military contexts.
• Current selection methods often lack predictive performance reporting or use uninterpretable 'black box' models.

Purpose of the Study:

• To introduce and evaluate a novel machine learning approach for selection research.
• To compare predictive performance, explainability, and stability of different machine learning models for identifying potential dropouts.

Main Methods:

• Analysis of data from 274 special forces recruits, including 196 who dropped out.
• Comparison of four machine learning models on physical and psychological test data.
• Utilized the Stable, Interpretable, and Robust, Understandable, Scalable (SIRUS) rule-based model.

Main Results:

• The SIRUS model demonstrated suitability for classifying special forces dropouts with an average Area Under the Curve (AUC) of 0.70.
• The model achieved good predictive performance while maintaining explainability and stability.
• Physical metrics (2800m run time, skin folds) and psychological factors (need for connectedness) were significantly associated with dropout.

Conclusions:

• The SIRUS model offers a valuable, explainable tool for selection research and practice.
• Insights into key predictive variables can inform interventions and improve selection strategies.
• This approach can enhance decision-making in high-performance environments like sports and military units.