Predicting a Fall Based on Gait Anomaly Detection: A Comparative Study of Wrist-Worn Three-Axis and Mobile...

Area of Science:

• Biomedical Engineering
• Gerontology
• Machine Learning

Background:

• Falls in the elderly present significant health risks, necessitating proactive fall prevention strategies.
• Early detection of gait deterioration can serve as a crucial indicator of impending falls.
• Wearable sensor technology offers potential for continuous monitoring and timely alerts.

Purpose of the Study:

• To compare the efficacy of a mobile phone system and two wristband systems (one commercial, one novel) in detecting fall-indicative gait changes.
• To evaluate the performance of machine learning techniques for gait analysis using accelerometer data from different wearable devices.
• To assess the potential of wearable sensors for early fall detection and mitigation in the elderly population.

Main Methods:

• A comparative analysis involving a smartphone, a commercial wristband, and a novel wristband, each with a three-axis accelerometer.
• Inducing fall-suggestive walking patterns using specialized glasses worn by participants.
• Applying standard machine learning techniques, including Support Vector Machine (SVM), for gait classification across all systems.
• Utilizing unsupervised and semi-supervised learning methods like Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (t-SNE).

Main Results:

• The novel wristband achieved the highest performance with a best average accuracy of 86%, specificity of 88%, and sensitivity of 86% using SVM.
• The smartphone system achieved a best average accuracy of 73% with SVM.
• Statistical analysis showed a significant difference (p-value < 0.000) in performance between the novel wristband and the smartphone.
• Both wristband systems demonstrated superior performance compared to the smartphone in detecting fall-indicative gait abnormalities.

Conclusions:

• Wearable wristbands equipped with accelerometers are effective tools for the early detection and mitigation of falls in the elderly.
• The novel wristband system demonstrated superior accuracy and sensitivity in identifying fall-risk gait compared to a smartphone.
• This study highlights the potential of integrating wearable technology and machine learning for enhancing elderly safety and preventing falls.