Abstract
This work proposes the use of bandstop filtering (BSF) as a pretreatment method in the quantitative analysis of glucose from both near-infrared (NIR) and midinfrared (MIR) spectra. The proposed method is investigated and evaluated against the traditional bandpass filtering (BPF) and implemented with the linear calibration models principal component regression (PCR) and partial least squares regression (PLSR) to predict the glucose from an aqueous mixture consisting of glucose and human serum albumin dissolved in a phosphate buffer solution. The results obtained show that BSF pretreatment achieves better prediction performance than BPF in both the NIR and MIR spectral regions. For detailed analysis, the BPF and BSF were implemented under both the Butterworth and Chebyshev filter configurations in both bands; in the NIR region, the Butterworth BSF combined with the PLSR model provides the best glucose prediction by reducing the root mean square error of prediction (RMSEP) from 100 mg/dL without filtering to 34 mg/dL with a coefficient of determination R-2 of .982. In the MIR region, the Chebyshev BSF combined with either PLSR or PCR improves the glucose prediction by reducing the RMSEP by 54% compared with 45% when using BPF and with R-2 of.995.