Effect of Day-To-Day Noise on UV-Visible Spectrophotometric Control Analyses of Mixtures by Principal Component Regression

Abstract

It is demonstrated that noise in UV spectral recordings obtained by using a diode array UV-visible spectrophotometer on different days may conform to a defined pattern. Such structured noise leads to the acceptance, as significant, of components containing noise alone, in calibrations by principal component regression (PCR)—which impedes the detection of outliers at the unknown sample prediction stage and considerably diminishes the potential of this methodology for control analyses. As shown in this paper, the effect of the noise structure can be substantially decreased by recording the spectra for the calibration samples on different days. Also, a procedure for distinguishing between correct samples and outliers is proposed. The procedure fits the distribution of the squared residuals of the absorbances for the calibration samples to an exponential function and uses a 99.9% probability as the acceptable limit. It was applied to analysis of ketoprofen and methylparaben mixtures.