Comparison of four machine learning algorithms for a pre-impact fall detection system.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Machine Learning for Healthcare

Background:

• Real-time health monitoring via wearable sensors is a significant research area.
• Fall detection systems are crucial for preventing injuries, especially in vulnerable populations.
• Existing systems often lack efficiency, affordability, or pre-impact detection capabilities.

Purpose of the Study:

• To develop and evaluate an efficient, low-cost fall detection system.
• To compare the performance of four machine learning algorithms for pre-impact fall detection.
• To assess the system's potential for providing timely intervention before a fall occurs.

Main Methods:

• A fall detection system was designed using shoes equipped with inertial and plantar pressure sensors.
• Four machine learning algorithms (K-Nearest Neighbors, Support Vector Machine, Random Forest, and BP neural network) were implemented and compared.
• Performance metrics including sensitivity, specificity, and accuracy were used to evaluate the algorithms for pre-impact detection.

Main Results:

• The K-Nearest Neighbors (KNN) and BP neural network algorithms demonstrated superior performance compared to SVM and Random Forest.
• KNN achieved 98.8% sensitivity, 99.8% specificity, and 99.7% accuracy.
• BP neural network achieved 100% sensitivity, 99.8% specificity, and 99.9% accuracy, with both algorithms providing a lead time of 460.95 ms for pre-impact detection.

Conclusions:

• The developed shoe-based system offers an effective and affordable solution for real-time fall detection.
• KNN and BP neural network algorithms are highly suitable for pre-impact fall detection, enabling timely intervention.
• The system has the potential to significantly reduce fall-related injuries when combined with protective devices.