Impact of Sensor Axis Combinations on Machine Learning Accuracy for Human Activity Recognition Using Accelerometer Data in Clinical Settings
Preprint
- 6 December 2024
- preprint
- Published by MDPI AG in Preprints.org
Abstract
For diseases such as chronic obstructive pulmonary disease (COPD) or arrhythmia, identifying the patient’s activity during symptom onset is crucial for severity assessment and diagnosis. Acceleration data from accelerometers can facilitate this process, but, in clinical settings, minimal data requirements are desirable. The study aimed to determine minimal number of axes needed from a 9-axis accelerometer worn on the non-dominant wrist or chest to maintain high accuracy in recognizing activities related to COPD or arrhythmia. Thirty healthy participants wore 9-axis accelerometers at five positions on their bodies and performed nine activities. We conducted machine learning-based activity recognition using 9-axis, 6-axis, and 3-axis data from the non-dominant wrist or chest. For the non-dominant wrist, 3-axis acceleration data recognized activities such as lying in the supine position, standing, taking a meal, and running, with accuracy comparable to 9-axis data. For the chest, 6-axis combining acceleration and magnetometer recognized activities such as lying in the supine position, standing, sitting, using the restroom, and ascending or descending stairs, with accuracy comparable to 9-axis data. These findings suggest that reducing the axes to 3-axis acceleration for the nondominant wrist and 6-axis combining acceleration and magnetometer for the chest, maintain high activity recognition accuracy.Keywords
All Related Versions
- Version 1, 2024-12-06, Preprints
- Published version: Physical Activity and Health, 9 (1).
This publication has 26 references indexed in Scilit:
- Investigating the impact of sensor axis combinations on activity recognition and fall detection: an empirical studyMultimedia Tools and Applications, 2024
- Assessing the impact of body location on the accuracy of detecting daily activities with accelerometer dataiScience, 2024
- A Comparative Study on Classifying Human Activities Using Classical Machine and Deep Learning MethodsArabian Journal for Science and Engineering, 2021
- Human activity recognition using magnetic induction-based motion signals and deep recurrent neural networksNature Communications, 2020
- Human Activity Recognition Using Smartphone Sensors: A Dense Neural Network Approach2019
- Machine learning CICY threefoldsPhysics Letters B, 2018
- Machine learning for activity recognition: hip versus wrist dataPhysiological Measurement, 2014
- Avoiding non-independence in fMRI data analysis: Leave one subject outNeuroImage, 2010
- Validation of a two-axis accelerometer for monitoring patient activity during blood pressure or ECG holter monitoringBlood Pressure Monitoring, 2003
- A readers' guide to the interpretation of diagnostic test properties: clinical example of sepsisIntensive Care Medicine, 2003