Non-invasive blood glucose estimation using Near-Infrared spectroscopy based on SVR

Abstract

There is a nonlinear relation between the blood glucose and photoplethysmography(PPG) signal. In order to estimate the blood glucose from the photoplethysmography signal, this paper presents a non-invasive blood glucose estimation using Near-Infrared spectroscopy based on the Support Vector Regression(SVR). The wavelet transform algorithm is used to remove baseline drift and smooth signals. 22 parameters, including features obtained from PPG signal and some physiological and environmental parameters, are the input parameters of Support Vector Regression model. The comparison between estimated and reference values shows better accuracy than the multiple linear regression analysis method, partial least squares method.