An Efficient Design of a Machine Learning-Based Elderly Fall Detector

Abstract

Elderly fall detection is an important health care application as falls represent the major reason of injuries. An efficient design of a machine learning-based wearable fall detection system is proposed in this paper. The proposed system depends only on a 3-axial accelerometer to capture the elderly motion. As the power consumption is proportional to the sampling frequency, the performance of the proposed fall detector is analyzed as a function of this frequency in order to determine the best trade-off between performance and power consumption. Thanks to efficient extracted features, the proposed system achieves a sensitivity of 99.73% and a specificity of 97.7% using a 40 Hz sampling frequency notably outperforming reference algorithms when tested on a large dataset.