BROWSE
Detail
Conference Paper
Fall Detection based on Interpretation of Important Features with Wrist-Wearable Sensors
Cited 5 time in 
Share 
Share
- Authors
- Jeong-Kyun Kim, Da-Som Oh, Kangbok Lee, and Sang Gi Hong
- Issue Date
- 2022-10
- Citation
- International Conference on Mobile Computing and Networking (MobiCom) 2022, pp.823-825
- Language
- English
- Type
- Conference Paper
- Abstract
- Smartwatches are attracting great industrial attention, and the demand for daily monitoring is increasing. Thus, a method for high fall detection with the smallest data is required to minimize the load on telecommunication. It is difficult to run machine learning on a smartwatch for fall detection, and the data must be transmitted to a mobile phone. Therefore, we propose a method for detecting a fall with few features by detecting important features in the inertial signal acquired from the wristband sensor. We used the FallAllD dataset for evaluation and detected 60 descriptive statistical features from each of the three-Axis acceleration and angular velocity. Additionally, important features were detected through SHapley additive exPlanations. The sampling frequency of the inertial signal obtained the highest accuracy of 94% at 120 Hz, with at least 14 important features.
- KSP Keywords
- 20 Hz, Angular Velocity, Daily monitoring, Fall Detection, Sampling frequency, Statistical Features, Three-axis acceleration, Wearable sensors, machine Learning, mobile phone