Heterogeneous Recognition of Human Activity with CNN and RNN-based Networks using Smartphone and Smartwatch Sensors
Issued Date
2022-01-01
Resource Type
Scopus ID
2-s2.0-85141549215
Journal Title
2022 3rd International Conference on Big Data Analytics and Practices, IBDAP 2022
Start Page
21
End Page
26
Rights Holder(s)
SCOPUS
Bibliographic Citation
2022 3rd International Conference on Big Data Analytics and Practices, IBDAP 2022 (2022) , 21-26
Suggested Citation
Mekruksavanich S., Jantawong P., Hnoohom N., Jitpattanakul A. Heterogeneous Recognition of Human Activity with CNN and RNN-based Networks using Smartphone and Smartwatch Sensors. 2022 3rd International Conference on Big Data Analytics and Practices, IBDAP 2022 (2022) , 21-26. 26. doi:10.1109/IBDAP55587.2022.9907460 Retrieved from: https://repository.li.mahidol.ac.th/handle/20.500.14594/84351
Title
Heterogeneous Recognition of Human Activity with CNN and RNN-based Networks using Smartphone and Smartwatch Sensors
Author's Affiliation
Other Contributor(s)
Abstract
With ongoing advancements in remote sensor technologies, the concept of wearable sensors has gained widespread acceptance and grown ubiquitous due to its broad applicability in situations including ambient assisted living, innovative healthcare, and home appliances. In this context, the demand for Heterogeneous Human Activity Recognition (HHAR) has expanded considerably, as individuals desire their daily activities to be collected and analyzed to provide valuable information. This wearable sensing data is taken in the form of time-series data utilizing different wearable sensors, such as accelerometers, gyroscopes, and magnetometers, to offer other movement characteristics for each individual tracked from their smartphones and smartwatches. This data may be analyzed using different deep learning architectures to provide an additional platform for capturing and classifying the numerous sensor-based actions an individual could be undertaking. This research compares HHAR models based on deep learning techniques. We study the influence of smartphone and wristwatch sensor data on model performance and classification of human activities. We train and evaluate deep learning models to achieve our study objective using a publically available standard dataset (HHAR) with a 5-fold cross-validation technique. Using smartphone sensors, research findings demonstrate that the RNN-based model is superior to CNN-based models, with the most fantastic accuracy of 98.42%.