TY - GEN
T1 - Using Deep Learning and Smartphone for Automatic Detection of Fall and Daily Activities
AU - Wu, Xiaodan
AU - Cheng, Lingyu
AU - Chu, Chao Hsien
AU - Kim, Jungyoon
N1 - Funding Information:
Acknowledgement. This project was supported in part by the National Social Science Foundation of China (No. 17BGL087). Our deepest gratitude goes to the anonymous reviewers for their careful review, comments and thoughtful suggestions that have helped improve this paper substantially.
Publisher Copyright:
© 2019, Springer Nature Switzerland AG.
PY - 2019
Y1 - 2019
N2 - The rapid growth of elderly population makes the health of the elderly one of the major social concerns. The elderly is often facing with several physical and mental healthcare related problems, among those, instance of fall and injuries ranked at the top. If people fall unexpectedly and without timely assistance, it is easy to cause irreparable harm. Therefore, how to automatically detect fall and alert for care/attention using advanced assisted technologies is a hot area of research. In this paper, we examine six machine learning-based methods and propose and carefully configure two novel deep learning-based architectures for fall detection. We compare the relative performance of these methods using an open source dataset, MobiAct, which was collected with four simulated fall types and nine daily living activities using smartphones. Our experimental results show that the proposed long short-term memory (LSTM) deep learning model is quite effective for the fall detection classification; its accuracy reaches 98.83%, the specificity is 99.38%, the sensitivity is 90.57% and the F1 score is 90.33%. These results are better than existing machine learning methods in all types of fall and most of daily activities.
AB - The rapid growth of elderly population makes the health of the elderly one of the major social concerns. The elderly is often facing with several physical and mental healthcare related problems, among those, instance of fall and injuries ranked at the top. If people fall unexpectedly and without timely assistance, it is easy to cause irreparable harm. Therefore, how to automatically detect fall and alert for care/attention using advanced assisted technologies is a hot area of research. In this paper, we examine six machine learning-based methods and propose and carefully configure two novel deep learning-based architectures for fall detection. We compare the relative performance of these methods using an open source dataset, MobiAct, which was collected with four simulated fall types and nine daily living activities using smartphones. Our experimental results show that the proposed long short-term memory (LSTM) deep learning model is quite effective for the fall detection classification; its accuracy reaches 98.83%, the specificity is 99.38%, the sensitivity is 90.57% and the F1 score is 90.33%. These results are better than existing machine learning methods in all types of fall and most of daily activities.
UR - http://www.scopus.com/inward/record.url?scp=85076751764&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85076751764&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-34482-5_6
DO - 10.1007/978-3-030-34482-5_6
M3 - Conference contribution
AN - SCOPUS:85076751764
SN - 9783030344818
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 61
EP - 74
BT - Smart Health - International Conference, ICSH 2019, Proceedings
A2 - Chen, Hsinchun
A2 - Zeng, Daniel
A2 - Yan, Xiangbin
A2 - Xing, Chunxiao
PB - Springer
T2 - 7th International Conference for Smart Health, ICSH 2019
Y2 - 1 July 2019 through 2 July 2019
ER -