Support Vector Machine Classifiers Show High Generalizability in Automatic Fall Detection in Older Adults.

Area of Science:

• Biomedical Engineering
• Machine Learning in Healthcare
• Gerontology

Background:

• Falls are a leading cause of death and disability, particularly in patients with neurological disorders.
• Accurate fall detection systems are crucial for timely intervention by caregivers and emergency services.
• A key challenge is ensuring machine learning models generalize from lab data to real-world scenarios.

Purpose of the Study:

• To evaluate the effectiveness of machine learning algorithms for automatic fall event detection.
• To assess the generalizability of fall detection classifiers across different datasets (laboratory vs. real-world).
• To identify the most suitable machine learning algorithm for reliable fall detection in practical applications.

Main Methods:

• Trained three machine learning algorithms: Support Vector Machine (SVM), k-Nearest Neighbors (kNN), and Random Forest (RF).
• Utilized the SisFall dataset (intentional falls) for classifier training.
• Validated the trained classifiers on the FARSEEING dataset (real-world accidental falls in elderly individuals).

Main Results:

• The linear SVM classifier demonstrated superior performance in cross-dataset validation.
• Linear SVM achieved 93% accuracy in distinguishing fall events from normal activities.
• The classifier exhibited similarly high sensitivity and specificity, indicating robust detection capabilities.

Conclusions:

• Machine learning classifiers, particularly linear SVM, show significant promise for automatic fall detection.
• The developed approach effectively generalizes from controlled to real-world fall events.
• Linear SVM-based systems could be valuable tools for enhancing safety and reducing fall-related morbidity and mortality.