Abstract
In recent years, Japan has been facing a serious aging population issue. Efforts are being made to utilize various sensors to enhance care quality in the field of caregiving to address this challenge. However, the use of sensors directly observing the caregiver′s state can be burdensome. In this study, we implemented a method to estimate the caregiver′s behavioral state using machine learning and environmental sensor data, which can be obtained without imposing a significant burden on the caregiver. Specifically, we utilized LSTM (long short-term memory), a machine learning technique capable of capturing the long-term time dependency of time-series data, to classify the caregiverbs state into three categories at each time point: sleeping, waking up, and going out. The proposed method was compared with an existing method that does not employ machine learning in terms of accuracy. It was confirmed that the proposed method outperforms the existing method. Finally, potential applications of the proposed method in the field of nursing care were discussed.
