Comparison of machine learning and deep learning models in manual strength prediction using anthropometric variables

Area of Science:

• Ergonomics and Biomechanics
• Machine Learning in Health Sciences

Background:

• Manual strength estimation is crucial for ergonomic assessments and occupational health.
• Anthropometric variables are commonly used predictors, but their efficacy with advanced models requires further investigation.

Purpose of the Study:

• To compare the predictive performance of various machine learning and deep learning models for estimating manual strength.
• To identify key anthropometric predictors influencing manual strength across different models.

Main Methods:

• Collected anthropometric and manual strength data from 382 participants.
• Implemented and evaluated linear regression, random forest, AdaBoost, extreme gradient boosting, TabNet, TabPFN, and a custom CNN.
• Assessed model performance using mean absolute error, mean squared error, and explained variance, with SHAP analysis for feature importance.

Main Results:

• Deep learning models (TabNet, TabPFN) showed superior accuracy for torque strength prediction.
• Linear regression demonstrated robust generalization, especially for grip strength.
• Palmar length and elbow-to-fingertip length were consistently identified as key predictors by SHAP analysis.

Conclusions:

• While advanced models improve specific strength predictions, linear regression provides better generalization.
• Model complexity should be balanced with interpretability for practical ergonomic applications.
• Anthropometric predictors like palmar length are biomechanically relevant for manual strength estimation.