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

OCCUPATIONAL APPLICATIONSThis study investigated whether tree-based machine learning algorithms can predict objective (EMG-based) and subjective (Borg CR10) low back fatigue during a sustained trunk flexion, with and without back-support exosuit assistance, using postural control features. The results revealed that all tree-based algorithms reasonably predicted objective and/or subjective fatigue during sustained trunk flexion (F1-score range: 74.1%-97.7%), with Extra Trees achieving the highest mean F1-score (91.4%). Fatigue detection was more accurate with subjective or combined fatigue labeling than with objective fatigue labeling alone, and exosuit assistance improved model precision and F1-score. Feature selection identified vertical peak and mean ground reaction forces as the most influential predictors. These results suggest that noninvasive postural control monitoring via smart insoles or force-sensing shoes, combined with machine learning, can enable real-time assessment of fatigue during trunk flexion, supporting timely ergonomic interventions and workload adjustments, or exoskeleton use to mitigate the risk of low back disorders.