Wavelength selection by net analyte signals calculated with multivariate factor-based hybrid linear analysis (HLA). A theoretical and experimental comparison with partial least-squares (PLS)

Abstract

A wavelength selection procedure is presented for the recently introduced factor-based multivariate calibration technique of hybrid linear analysis (HLA). It involves the calculation of the net analyte signal regression plot (NASRP) for each test sample, combined with a moving window strategy. A search for the minimum error indicator (EI) is performed using the first significant factors of a data matrix from which the contribution of the analyte of interest has been removed. HLA uses less factors than both principal component regression (PCR) and partial least-squares (PLS), and is simpler to adapt to the NASRP methodology. The ability of the method to select wavelength regions that minimize the effect of non-modeled interferences is illustrated both with simulated examples and with experimental bromhexine-containing cough suppressant syrup samples.