Abstract
In the present study, a dry film-based Fourier transformed-infrared (FT-IR) spectroscopic technique, coupled with boosting support vector regression (BSVR), was employed for a blood glucose assay. Potassium thiocyanate (KSCN) was taken in the dry-film method as an internal standard to compensate for any film thickness variation. This technique circumvents interference from water absorption, and requires only 5 µl of a sample. Moving window partial least-squares regression (MWPLSR) was used for wavenumber interval selection before multivariate modeling. By using the BSVR modeling technique, glucose in plasma could be determined over a 0.4 - 20 mmol/l concentration range with satisfactory accuracy. The performance of the BSVR methodology was compared with that of conventional support vector regression (SVR) as well as partial-least squares (PLS). The results demonstrated that BSVR is an effective multivariate calibration tool, providing better performance than conventional PLS and SVR.
