Article
A Class-Imbalanced Deep Learning Fall Detection Algorithm
Using Wearable Sensors
Jing Zhang
1,2
, Jia Li
1,2,
* and Weibing Wang
1,2
Citation: Zhang, J.; Li, J.; Wang, W. A
Class-Imbalanced Deep Learning Fall
Detection Algorithm Using Wearable
Sensors. Sensors 2021, 21, 6511.
https://doi.org/10.3390/
s21196511
Academic Editor: Nunzio Cennamo
Received: 25 August 2021
Accepted: 27 September 2021
Published: 29 September 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
1
Institute of Microelectronics of Chinese Academy of Sciences, Beijing 100029, China;
zhangjing@ime.ac.cn (J.Z.); wangweibing@ime.ac.cn (W.W.)
2
School of University of Chinese Academy of Sciences, Beijing 100049, China
* Correspondence: lijia@ime.ac.cn
Abstract:
Falling represents one of the most serious health risks for elderly people; it may cause
irreversible injuries if the individual cannot obtain timely treatment after the fall happens. Therefore,
timely and accurate fall detection algorithm research is extremely important. Recently, a number of
researchers have focused on fall detection and made many achievements, and most of the relevant
algorithm studies are based on ideal class-balanced datasets. However, in real-life applications, the
possibilities of Activities of Daily Life (ADL) and fall events are different, so the data collected by
wearable sensors suffers from class imbalance. The previously developed algorithms perform poorly
on class-imbalanced data. In order to solve this problem, this paper proposes an algorithm that can
effectively distinguish falls from a large amount of ADL signals. Compared with the state-of-the-art
fall detection algorithms, the proposed method can achieve the highest score in multiple evaluation
methods, with a sensitivity of 99.33%, a specificity of 91.86%, an F-Score of 98.44% and an AUC of
98.35%. The results prove that the proposed algorithm is effective on class-imbalanced data and more
suitable for real-life application compared to previous works.
Keywords: fall detection; class imbalance; deep learning; wearable sensor
1. Introduction
Falls have become the second largest threat to the health of people worldwide, and
most of the people who die from falls are over 60 years old. The elderly are at the greatest
risk of suffering from falls. Falls result in fatal injuries, such as paralysis, hip fracture,
head injury, etc. [
1
]. More than 37.3 million falls occur each year that are severe enough to
require medical attention [
2
]. With the development of society, the aging of the world’s
population is of concern. Moreover, the number of elderly people living alone is constantly
increasing. In the context of living alone, when an elderly individual falls, it is difficult for
them to seek help on their own without a fall detection system (FDS). If the individual does
not receive timely medical attention, there may be serious consequences. It is important to
rescue elderly individuals in a short time after a fall happens, and the application of FDS
has become very significant in this context.
Many researchers have focused on human fall detection and have carried out a great
deal of work. In the past decade, a large number of fall detection programs have been
proposed, which can be divided into the following three categories: video-based [
3
–
7
],
ambient sensor-based [
8
–
11
] and wearable sensor-based [
12
–
18
], as shown in Figure 1. The
sensitivity and the specificity of the video-based methods can reach 97% and 99%, respec-
tively [
4
], indicating that they can accurately identify the occurrence of falls. However,
visualization tools such as cameras used for video image detection are usually fixed, so
they are more suitable for indoor use. Moreover, these devices have limitations, such as
being easily blocked by objects, and cameras are not privacy-preserving. Furthermore,
ambient sensor-based methods are less disruptive to people’s lives, and the sensitivity can
reach 97% while the specificity can reach 95% [
8
]. Nevertheless, ambient sensor-based
Sensors 2021, 21, 6511. https://doi.org/10.3390/s21196511 https://www.mdpi.com/journal/sensors
