Predicting Maximal Military Occupational Task Performance from Physical Fitness Tests Using Machine Learning.

Area of Science:

• Military science
• Sports science
• Data science

Background:

• Military task performance is vital for operational success.
• Current physical assessments are time-consuming and pose injury risks.
• Predictive models using physical characteristics can optimize military personnel management.

Purpose of the Study:

• To develop machine learning models for predicting military physical assessment outcomes.
• To assess the accuracy of these models in predicting lift-to-place and jerry-can-carry tasks.
• To explore the potential of machine learning in optimizing military training and selection.

Main Methods:

• Recruited 35 participants for two physical assessment sessions.
• Collected data on multiple physical characteristics and performance in lift-to-place and jerry-can-carry tasks.
• Developed and evaluated machine learning models (SVR, Ridge, Multi-Layer Perceptron) using RMSE, NRMSE, and CVRMSE.

Main Results:

• Support Vector Regression and Ridge models accurately predicted lift-to-place outcomes with low error (RMSE ±1.77 kg and ±2.33 kg).
• Multi-Layer Perceptron and SVR models showed promise for predicting jerry-can-carry outcomes (RMSE ±3.36 laps and ±3.67 laps).
• Model accuracy varied, with lift-to-place predictions being more precise than jerry-can-carry predictions.

Conclusions:

• Machine learning models can accurately predict certain military physical task outcomes, like lift-to-place.
• Further optimization is needed for predicting tasks like the jerry-can-carry.
• These models offer a promising approach to optimize military occupational selection and training interventions.