Research Article
ADLs Detection with a Wrist-Worn Accelerometer in Uncontrolled Conditions
@INPROCEEDINGS{10.1007/978-3-030-76066-3_16, author={Sandro Fioretti and Marica Olivastrelli and Angelica Poli and Susanna Spinsante and Annachiara Strazza}, title={ADLs Detection with a Wrist-Worn Accelerometer in Uncontrolled Conditions}, proceedings={Wearables in Healthcare. Second EAI International Conference, HealthWear 2020, Virtual Event, December 10-11, 2020, Proceedings}, proceedings_a={HEALTHWEAR}, year={2021}, month={5}, keywords={Human activity recognition Wearable sensors Machine learning classifiers}, doi={10.1007/978-3-030-76066-3_16} }- Sandro Fioretti
Marica Olivastrelli
Angelica Poli
Susanna Spinsante
Annachiara Strazza
Year: 2021
ADLs Detection with a Wrist-Worn Accelerometer in Uncontrolled Conditions
HEALTHWEAR
Springer
DOI: 10.1007/978-3-030-76066-3_16
Abstract
In 2017, the European Commission estimated that 29% of European population will be aged 65 and over, by 2070. The capability of tracking and recognizing people’s daily activities may promote and support an active and independent lifestyle. In this regard, Human Activity Recognition allows to obtain meaningful information by monitoring daily activities using wearable devices, that are small, easy to use, and minimally invasive.
In this paper, we discuss the recognition performance of six machine learning classifiers applied to accelerometer data only. Data was collected by 36 individuals, wearing a single wrist-worn sensor to monitor six daily activities pertaining to Hygiene and House Cleaning scenarios. Following a pre-processing phase, both temporal and frequency features were computed to classify and recognize the collected real-world data. The study presents some statistical results obtained from each classifier in order to compare their performance. The findings of experiments are promising for the adoption of the Random Forest classifier in Human Activity Recognition with acceleration data from a single wrist-worn device.
