In the last few decades, life expectancy has been increasing. This has resulted in a higher proportion of older adults and increased prevalence of chronic conditions, posing challenges facing care needs. A possible solution is to foster both the prevention and health-related re-education, supporting healthier lifestyle and facilitating independent living. To facilitate this, it is crucial to measure individual’s key health metrics. For instance, human activity recognition through sensors provides valuable information about an individual’s lifestyle. Some crucial decisions, among which the quality of data collection, strengthen the methodological approach. This chapter addresses how the quality of data may affect the recognition performance. Two datasets of daily activities were collected through a triaxial accelerometer placed on the subject’s dominant wrist. The first dataset was collected by 141 users, whereas the second one comprised semi-realistic activities executed by three individuals. Specifically, outcomes were based on a comparison of activity recognition performance of six machine learning classifiers. Results show that, firstly, a higher number of features may not improve the recognition rate. Secondly, one approach may be robust in a laboratory setting but not generalizable to real-world applications. Finally, a great variability may increase the generalization of classifiers for successful activity recognition.
Improving the Collection and Understanding the Quality of Datasets for the Aim of Human Activity Recognition / Poli, Angelica; Spinsante, Susanna; Nugent, Chris; Cleland, Ian. - ELETTRONICO. - (2019), pp. 147-165. [10.1007/978-3-030-25590-9_7]
Improving the Collection and Understanding the Quality of Datasets for the Aim of Human Activity Recognition
Poli, Angelica
Writing – Original Draft Preparation
;Spinsante, SusannaWriting – Review & Editing
;
2019-01-01
Abstract
In the last few decades, life expectancy has been increasing. This has resulted in a higher proportion of older adults and increased prevalence of chronic conditions, posing challenges facing care needs. A possible solution is to foster both the prevention and health-related re-education, supporting healthier lifestyle and facilitating independent living. To facilitate this, it is crucial to measure individual’s key health metrics. For instance, human activity recognition through sensors provides valuable information about an individual’s lifestyle. Some crucial decisions, among which the quality of data collection, strengthen the methodological approach. This chapter addresses how the quality of data may affect the recognition performance. Two datasets of daily activities were collected through a triaxial accelerometer placed on the subject’s dominant wrist. The first dataset was collected by 141 users, whereas the second one comprised semi-realistic activities executed by three individuals. Specifically, outcomes were based on a comparison of activity recognition performance of six machine learning classifiers. Results show that, firstly, a higher number of features may not improve the recognition rate. Secondly, one approach may be robust in a laboratory setting but not generalizable to real-world applications. Finally, a great variability may increase the generalization of classifiers for successful activity recognition.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.