A Movement Decomposition and Machine Learning-Based Fall Detection System Using Wrist Wearable Device

Falls in the elderly is a world health problem. Although many fall detection solutions were presented in literature, few of them are wrist-wearable devices, mainly due to typical processing and classification challenges to achieve accuracy greater than 95%. Considering the wrist as a more comfortable, discrete and acceptable place for an elderly wearable device, this paper presents the development and evaluation of a wrist-worn fall detection solution. Different sensors (accelerometer, gyroscope, and magnetometer), signals (acceleration, velocity, and displacement), and direction components (vertical and non-vertical) were combined and a comprehensive set of threshold-based and machine learning methods were applied in order to define the best approach for fall detection. Data was acquired for fall and non-fall movements from 22 volunteers. For threshold-based methods, a maximum accuracy of 91.1% was achieved with 95.8% and 86.5% of sensitivity and specificity, respectively, using Madgwick's decomposition. With the same movement decomposition and machine learning methods in the classification stage, an impressive accuracy of 99.0% was achieved, with 100% of sensitivity and 97.9% of specificity in our data set. Prolonged tests with a volunteer wearing the fall detector also demonstrate the advantages of machine learning methods in terms of practical applications.