Non-invasive blood glucose detection using NIR based on GA and SVR

Between the human blood glucose with photoplethysmography (PPG) signal, there is a nonlinear and complex relation. For the purpose of measuring the human blood glucose level (BGL) from PPG signal, in this paper, we propose one non-invasive blood glucose detection method using NIR based on Genetic Algorithm (GA) and Support Vector Regression (SVR). During data preprocessing, we use the wavelet transform algorithm to smooth the PPG signal and eliminate baseline drift. 23-dimensional parameters, including some physiological and environmental parameters and features calculated from PPG signal, compose the feature matrix. Genetic Algorithm is used for feature selection. Selected parameters are used for SVR model training. On the test set, we used the method based on GA and SVR to obtain a 95.05% correlation coefficient, 1.74% higher than the method using SVR only.