Abstract

Info

Title:: Application of co-training algorithm in noninvasive blood glucose detection

Author(s):: ZHANG Di1; CHEN Zhencheng2; LIANG Yongbo2; WU Zhiqiang3; ZHU Jianming2; ZHONG Tingting1; 1. School of Electronic Engineering and Automation, Guilin University of Electronic Technology, Guilin 541004, China; 2. School of Life and Environmental Sciences, Guilin University of Electronic Technology, Guilin 541004, China; 3. Medical Devices Testing Center of Guangxi Zhuang Autonomous Region, Nanning 530021, China

Keywords:: ; Keywords: database; noninvasive blood glucose detection; co-training; support vector machine

Abstract:: Abstract: Objective In the study of noninvasive blood glucose detection, obtaining noninvasive physiological parameters is easier than obtaining true blood glucose. In the pathological database, the number of samples unlabeled with true blood glucose is much larger than that of labeled samples. This research aims to apply the unlabeled samples to the training of traditional supervised model for the prediction of blood glucose for effectively expanding the training set and improving the generalization ability of the model. Methods Based on the theory of conservation of energy metabolism and the natural multi-view characteristics of noninvasive physiological parameters, a semi-supervised learning algorithm was applied to the prediction of blood glucose. An algorithm based on multi-view co-training and support vector regression was proposed for the prediction of blood glucose. Results The experimental analysis showed that at a certain labeling rate, the prediction error of algorithm based on co-training is lower than that of traditional supervised learning algorithm, which indicated that unlabeled samples could effectively improve the generalization ability of the original model. Conclusion Co-training algorithm can fully utilize large-scale unlabeled samples, improve the generalization ability of model, and reduce the workload of labeling blood glucose samples, providing a new idea for the future research on noninvasive blood glucose algorithm.