Clustering Analysis of Activity of Daily Living based on Accelerometry

Abstract

This study aims to categorize individual Activities of Daily Living (ADL) into activity categories by utilizing acceleration features. The research involves extracting time and frequency features from 10 different types of ADL acceleration data. These features are used to cluster each behavior through a bottom-up clustering approach, specifically the Ward Linkage method. The findings of this study reveal that bottom-up clustering, utilizing time-domain and frequency-domain acceleration features, provides insights into behavior characteristics such as magnitude, frequency, and repeatability, as indicated by metrics like acceleration magnitude and frequency distribution. This approach allows for the clustering of relatively similar and interconnected behaviors. Notably, various time and frequency domain characteristics, including magnitude per unit time and concentration at specific frequencies, were identified as effective indicators for behavior clustering, in addition to the commonly used total acceleration magnitude.