Importance of Personalized Health-Care Models: A Case Study in Activity Recognition
Journal
Studies in health technology and informatics
Date Issued
2018-01-01
Author(s)
Pombo, Nuno
Garcia, Nuno
Abstract
Novel information and communication technologies create possibilities
to change the future of health care. Ambient Assisted Living (AAL) is seen as a
promising supplement of the current care models. The main goal of AAL solutions
is to apply ambient intelligence technologies to enable elderly people to continue
to live in their preferred environments. Applying trained models from health data
is challenging because the personalized environments could differ significantly
than the ones which provided training data. This paper investigates the effects on
activity recognition accuracy using single accelerometer of personalized models
compared to models built on general population. In addition, we propose a
collaborative filtering based approach which provides balance between fully
personalized models and generic models. The results show that the accuracy could
be improved to 95% with fully personalized models, and up to 91.6% with
collaborative filtering based models, which is significantly better than common
models that exhibit accuracy of 85.1%. The collaborative filtering approach seems
to provide highly personalized models with substantial accuracy, while
overcoming the cold start problem that is common for fully personalized models.
to change the future of health care. Ambient Assisted Living (AAL) is seen as a
promising supplement of the current care models. The main goal of AAL solutions
is to apply ambient intelligence technologies to enable elderly people to continue
to live in their preferred environments. Applying trained models from health data
is challenging because the personalized environments could differ significantly
than the ones which provided training data. This paper investigates the effects on
activity recognition accuracy using single accelerometer of personalized models
compared to models built on general population. In addition, we propose a
collaborative filtering based approach which provides balance between fully
personalized models and generic models. The results show that the accuracy could
be improved to 95% with fully personalized models, and up to 91.6% with
collaborative filtering based models, which is significantly better than common
models that exhibit accuracy of 85.1%. The collaborative filtering approach seems
to provide highly personalized models with substantial accuracy, while
overcoming the cold start problem that is common for fully personalized models.
Subjects
File(s)![Thumbnail Image]()
Loading...
Name
2018_pHealth_revised_submission3_Zdravevski_final.pdf
Size
185.48 KB
Format
Adobe PDF
Checksum
(MD5):cce62c03904b49492b28014ab5ae7a0f