Logo image
Back
A machine learning approach to identify fall risk for older adults
Journal article   Peer reviewed

A machine learning approach to identify fall risk for older adults

Prithwi Raj Chakraborty and Golam Sorwar
Smart Health, Vol.26, 100303
12/2022
DOI: https://doi.org/10.1016/j.smhl.2022.100303
url
A machine learning approach to identify fall risk for older adults View
Published (Version of record)

Related links

Metrics

80 Record Views

Abstract

Accelerometer data Fall risk identification Machine learning approach Statistical analysis Machine learning not elsewhere classified Digital health Artificial intelligence Health surveillance
Fall risk identification can save lives and health service expenses for older adults. This work has attempted to introduce a new method using statistical and machine learning approaches to automatically identify fall risks for older adults. Several time-domain features from long-term accelerometer data are extracted for experimentation. Preliminary correlation analysis found no significant correlation between age and the number of falls. Statistical analysis is conducted to confirm the significant statistical difference between the distributions of the “faller” and “non-faller” data in overall 82%–96% occurrences. This work adopts a classification-based approach to identify fall risk, considering that a person with a previous history of falls has a higher risk of future falls. Three classification approaches (Approach 1, 2, and 3) have been used along with two cross-validations that achieved an overall accuracy between 58%–78%, 73%–94%, and 76%–96% respectively. Approach 1 does not use any data shuffling and feature selection, approach 2 uses data shuffling without any feature selection, and approach 3 uses both data shuffling and feature selection. Experimental results show that data shuffling and feature selection effectively increase classification performance.

Details

Logo image