Predicting manual arm strength: A direct comparison between artificial neural network and multiple regression...

Area of Science:

• Biomechanics
• Ergonomics
• Occupational Health

Background:

• Traditional strength prediction in ergonomics relies on linked-segment biomechanical models.
• Multiple regression models show promise for predicting manual arm strength (MAS) using hand location and force direction.
• Application of artificial neural networks (ANNs) in occupational biomechanics and ergonomics is currently limited.

Purpose of the Study:

• To directly compare the predictive accuracy of artificial neural networks (ANNs) and regression models for manual arm strength (MAS).
• To evaluate the performance of ANNs and regression models using identical development and validation datasets.

Main Methods:

• Collected multi-directional MAS data from 95 healthy female participants across 36 hand locations.
• Developed ANN and regression models using 85% of the MAS data (n=456).
• Validated both models using the remaining 15% of the data (n=80).

Main Results:

• ANN models demonstrated significantly higher explained variance (90.2%) and lower RMSD (9.3N) on development data compared to regression models (66.5%, 17.2N).
• ANNs outperformed regression on independent validation data, showing higher r² (78.6% vs. 65.3%) and lower RMSD (15.1N vs. 18.6N).

Conclusions:

• Artificial neural networks (ANNs) provide a more accurate and robust prediction of manual arm strength (MAS) than regression models.
• ANNs should be more frequently considered for biomechanics and ergonomics evaluations.
• This study highlights the potential of ANNs to advance strength prediction in occupational settings.