Accelerometer-Based Fall Detection Using Machine Learning: Training and Testing on Real-World Falls.

Area of Science:

• Biomedical Engineering
• Gerontology
• Machine Learning

Background:

• Falls pose a significant health risk, particularly for individuals with moderate-to-high fall risk.
• Current fall detection systems often rely on simulated falls, limiting real-world applicability.
• Effective fall detection is crucial for timely medical intervention and improved patient outcomes.

Purpose of the Study:

• To analyze real-world fall data to develop and test advanced fall detection algorithms.
• To compare the performance of algorithms using multiphase fall model features against conventional features.
• To establish reliable metrics for characterizing real-world fall detection systems.

Main Methods:

• Analysis of acceleration signals from inertial sensors during 143 real-world falls from the FARSEEING repository.
• Development of fall detection algorithms incorporating features inspired by a multiphase fall model.
• Application of a machine learning approach, specifically support vector machines, for classification.

Main Results:

• Algorithms leveraging multiphase fall model features consistently outperformed conventional features.
• The most effective method achieved over 80% sensitivity and a false alarm rate of 0.56 per hour.
• An F-measure of 64.6% was obtained, indicating a strong performance in real-world fall detection.

Conclusions:

• Machine learning algorithms can effectively learn from multiphase fall model features for improved fall detection.
• The developed methods and metrics advance the understanding of real-world fall detection systems.
• This research provides a robust framework for evaluating fall detection technologies in practical settings.