Improving Accuracy of Inkjet Printed Core Body WRAP Temperature Sensor Using Random Forest Regression Implemented with an Android App

Abstract

Inkjet printing (IJP) technology holds tremendous promise for the development of low cost, environment friendly and body-worn biomedical sensors. In this study, we have investigated the integration of a flexible body-worn disposable IJP Wireless Resistive Analog Passive (WRAP) temperature sensor with an android app for real-time monitoring of core body temperature with high accuracy using features extracted from the sensor response. Random Forest has been used for feature selection and regression. With 5-fold cross validation we have achieved an RMSE = 0.98, R-squared value = 0.99, and mean absolute error, MAE = 0.59 for temperature estimation. The model is applicable for the development of IJP body-worn sensors for various other physiological sensing e.g. breathing, heart rate.