Human body skin temperature prediction based on machine learning

Abstract

Body temperature is one of the vital indicators that reflects the human body health. It is used to diagnose different types of diseases. Generally, there are two types of body temperatures: the skin and core temperature. The body temperature is used to diagnose some diseases by assessing its abnormality. The core temperature is primarily used for diagnosis as a clear standard is available. However, core temperature measurement requires minutes and places a strain on the subjects, whereas the skin temperature can be monitored real time and measured without contact. Because the skin temperature is affected by many factors, a definite standard to assess its abnormality does not exist. Herein, we propose a prediction method for the skin temperature based on machine learning. Each estimator of the machine learning is generated in restricted conditions to limit the factors handled by machine learning at once. The root mean square error of the prediction is 0.315 ℃, and the root mean square percentage error is 0.90%. The prediction result can be used as a standard for the anomaly detection of the skin temperature.