Improving Noninvasive blood glucose measurement accuracy by applying genetic algorithm to partial least square regression model

Near infrared (NIR) absorption spectroscopy is a promising technique to noninvasively quantify blood glucose level. In order to extract the glucose signal out of the noisy background, Partial Least Squares (PLS) was utilized to create calibration models that relate the absorption spectra to glucose concentrations. A research grade Fourier Transformed Infrared (FTIR) spectrometer configured with a NIR quartz beam-splitter was used in this investigation. Genetic Algorithm (GA) was implemented to search the most appropriate modeling parameters such as wavelengths within NIR range for PLS regression. Using GA method to optimize the wavelength selection by applying the PLS-based calibration model could greatly enhance the prediction capacity and improve the measurement accuracy.