Evaluation of Feature Engineering on Wearable Sensor-based Fall Detection

Abstract

Internet of Things (IoT) enabled geriatric healthcare systems have gained importance in the recent years due to an increase in the number of elderly people living alone. The application of machine learning (ML) in areas of geriatric healthcare such as fall detection, has, consequently been an area of active research. Wearable systems for fall detection has the advantage of being light-weight and low power-consuming, yet reasonably accurate without being overly intrusive. However, the accuracy of fall detection using wearable systems depends, among other factors, on the types of sensors embedded in them. The use of inertial motion unit (IMU) sensors for fall detection, combined with machine learning classifiers applied on datasets collected from IMU sensors is an area of active research. In this paper, we analyze the impact of using IMU sensor parameters in combination with vital signs parameters, on the performance of ML algorithms for fall detection. We compare the performance of various ML algorithms on the dataset we collected for this purpose. We also perform statistical analysis to examine the relative importance of the various features on the behavior of ML classifiers.