Machine learning models classify walker usage and posture to improve fall prevention in older adults

Researchers evaluated multiple machine learning approaches—Geometric, XGBoost, SVM, and deep learning architectures—to classify walker usage, standing vs. sitting, and posture in smart walkers. XGBoost and the Geometric approach were top performers.

XGBoost achieved 99.84% training accuracy for walker choice classification and 99.69% for standing vs. sitting tasks. For posture classification across 17 postures, XGBoost obtained 99.24% training accuracy; the Geometric approach attained 89.9% accuracy for 8 postures. Deep learning models (4-layer CNN and Encoder-Decoder CNN) also demonstrated accuracies above 98% in binary classification.

The study underscores the potential of machine learning to enhance human-robot interaction in smart walkers, particularly for fall prevention—a significant public health concern among older adults that leads to severe injuries, loss of independence, and increased healthcare costs.