A Machine Learning Approach to Predict Objective and Subjective Low Back Fatigue Using Postural Control Features...

Area of Science:

• Occupational health and safety
• Biomechanical engineering
• Machine learning applications

Background:

• Low back fatigue is a significant occupational hazard, particularly during sustained trunk flexion.
• Current fatigue assessment methods are often subjective or invasive.
• Exoskeleton technology offers potential for mitigating physical strain in occupational settings.

Purpose of the Study:

• To evaluate tree-based machine learning algorithms for predicting objective and subjective low back fatigue.
• To assess the impact of back-support exosuit assistance on fatigue prediction accuracy.
• To identify key postural control features for fatigue detection.

Main Methods:

• Utilized tree-based algorithms (e.g., Extra Trees) to analyze postural control features during sustained trunk flexion.
• Incorporated objective (EMG) and subjective (Borg CR10) fatigue measures.
• Investigated the effect of exosuit assistance on model performance.
• Performed feature selection to identify critical predictors.

Main Results:

• All tested algorithms demonstrated reasonable to high accuracy (F1-scores 74.1%-97.7%) in predicting fatigue.
• Extra Trees algorithm yielded the highest mean F1-score (91.4%).
• Subjective or combined fatigue labeling improved prediction over objective labeling alone.
• Exosuit assistance enhanced model precision and F1-score.
• Vertical ground reaction forces were identified as key predictors.

Conclusions:

• Noninvasive postural control monitoring via smart insoles or force-sensing shoes, coupled with machine learning, can enable real-time fatigue assessment.
• This approach supports timely ergonomic interventions and workload adjustments.
• Exoskeleton use can be optimized based on real-time fatigue data to mitigate low back disorder risks.