Abstract
Invasive methods of core body temperature measurement pose challenges to the assessment and supervision of human health. Although non-invasive methods are available, they frequently involve the use of medical equipment. To address this issue, some methods have been proposed various types of biological data. However, existing systems often require large measurement devices and complex calculations, limiting their practicality. In this paper, we present an estimation model that leverages machine learning to predict core body temperature using a hand-mounted wearable sensor. In the preprocessing stage, we identified features based on the peak-to-peak interval derived from a volume pulse wave and created lagged datasets using biological data. To account for non-linearity in the machine learning component, we used support vector regression. Our proposed method displayed an error of 0.078 ℃ in core body temperature estimation. On the other hand, the application conditions of this model are limited, and it is necessary to acquire biological data considering various environments, such as hot and humid conditions or during physical activities.
