Movement Pattern Analysis Based on Point-Line-Plane Hierarchies and Machine Learning for Fall Risk Assessment in...

Area of Science:

• Gerontology
• Biomechanics
• Machine Learning

Background:

• Falls are a major health concern for community-dwelling older adults.
• Accurate fall risk classification is essential for timely intervention.
• Traditional methods often fail to capture the complexity of postural control.

Purpose of the Study:

• To compare the effectiveness of different body segment kinematic features for fall risk classification.
• To identify key kinematic variables predictive of fall risk.
• To evaluate the performance of machine learning models in fall risk prediction.

Main Methods:

• Systematic comparison of full-body, upper limb, lower limb, and trunk kinematic features.
• Application of machine learning models (XGBoost, random forest) for classification.
• Feature importance analysis to identify critical predictive variables.

Main Results:

• Full-body and upper limb features demonstrated higher accuracy in fall risk classification compared to lower limb and trunk.
• Selected key variables, including wrist displacement and upper limb velocity, boosted model accuracy from 57% to 78%.
• Machine learning models like XGBoost and random forest outperformed traditional linear models (max accuracy 0.61).

Conclusions:

• Upper limb movement plays a critical role in maintaining balance and predicting fall risk.
• Integrating multi-regional movement coordination enhances fall risk prediction accuracy.
• Advanced machine learning models and comprehensive kinematic data offer promising tools for clinical fall risk screening.