Domain-Adaptive Uncertainty-Aware Temporal Bagging Ensemble for Gait Phase Estimation

Key Points

• The aim is to develop a robust ensemble method for estimating continuous gait phases across different walking environments.
• Utilized wearable multi-sensors (IMU, FSR, soft bending sensor) for data collection.
• Developed an uncertainty-aware temporal bagging ensemble model (UA-TBE).
• Classified walking environments to assess environmental probabilities and domain uncertainties.
• Integrated predictions from sub-models trained on different time window ratios for adaptive ensemble weighting.
• UA-TBE outperformed simple average ensemble with a 69.8% improvement in MAE.
• Demonstrated a 71.1% reduction in RMSE.
• Achieved a 2.5% increase in R², indicating better predictive accuracy.
• Showed that the model suppressed unnecessary overconfidence in transitional environments.