Detecting falls with wearable sensors using machine learning techniques.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Machine Learning Applications

Background:

• Falls pose a significant public health risk, especially for vulnerable populations.
• Existing fall detection systems often rely on simple thresholding, limiting their accuracy.
• Automated systems are needed for reliable fall detection in real-world scenarios.

Purpose of the Study:

• To develop and evaluate an automated fall detection system using wearable motion sensors.
• To compare the performance of six different machine learning classifiers for fall detection.
• To assess the computational efficiency of the developed system.

Main Methods:

• Utilized wearable motion sensor units (accelerometer, gyroscope, magnetometer) at six body positions.
• Collected data from 14 volunteers performing voluntary falls and activities of daily living (ADLs).
• Applied feature extraction and reduction on a 4-second time window around peak acceleration, using six machine learning classifiers (k-NN, LSM, SVM, BDM, DTW, ANNs).

Main Results:

• Achieved high performance in distinguishing falls from ADLs with sensitivity, specificity, and accuracy exceeding 99% using k-NN and LSM classifiers.
• k-NN and LSM demonstrated acceptable computational complexity for training and testing.
• The system effectively processed raw sensor data for classification.

Conclusions:

• The developed automated fall detection system demonstrates high accuracy and efficiency.
• k-NN and LSM classifiers are highly effective for real-time fall detection applications.
• The system is suitable for real-world deployment in scenarios with continuous data recording.