Research on data augmentation for vital data using conditional GAN

• Published in: 2022 IEEE 11th Global Conference on Consumer Electronics (GCCE) pp. 344 - 345
• Main Authors: Shiotani, Maho, Iguchi, Shino, Yamaguchi, Katsuhisa
• Format: Conference Proceeding
• Language: English
• Published: IEEE 18.10.2022
• Subjects: data augmentation; IoT sensor; lifestyle data; vital data
• DOI: 10.1109/GCCE56475.2022.10014132

In the field of elderly care, detecting anomaly in condition of the elderly people is important for improving the quality of care and quality of life. However, anomaly detection technology using machine learning often requires a large amount of training data. In this study, we examined whether vital and lifestyle data can be extended appropriately using generative adversarial networks. As vital and lifestyle data, heart rate, respiration rate, and bed leaving status were collected from sheet-type sensors set on subject's bed. Data augmentation was conducted on four nursing home users. As a result, the difference between the average values of the actual data and the extended data for heart rate and respiratory rate was less than 10% of the average of the actual data, which indicates that the extended data was successfully generated relatively close to the actual data. On the other hand, individual differences were shown in the extended performance of the bed leaving status. Future work is needed to consider how to extend the lifestyle data in consideration of individual differences.