Abstract
Noising is an undesirable phenomenon accompanying the development of widely used chemometric models such as partial least square regression (PLSR) and support vector regression (SVR).
Optimizations of these chemometric models by applying orthogonal projection to latent structures (OPLS) as a preprocessing step which is characterized by canceling noise is the purpose of this research study. Additionally, a comprehensive comparative study between the developed methods was undertaken highlighting pros and cons.
OPLS was conducted with PLSR and SVR for quantitative determination of pyridoxine HCl, cyclizine HCl, and meclizine HCl in the presence of their related impurities. The training set was formed from 25 mixtures as there were five mixtures for each compound at each concentration level. Additionally, to check the validity and predictive ability of the developed chemometric models, independent test set mixtures were prepared by repeating the preparation of four mixtures of the training set plus preparation of another four independent mixtures.
Upon application of the OPLS processing method, an upswing of the predictive abilities of PLSR and SVR was found. The root-mean-square error of prediction of the test set was the basic benchmark for comparison.
The major finding from the conducted research is that processing with OPLS reinforces the ability of models to anticipate the future samples.
Novel optimizations of the widely used chemometric models; application of a comparative study between the suggested methods; application of OPLS preprocessing methods; quantitative determination of pyridoxine HCl, cyclizine HCl and meclizine HCl; checking the predictive power of developed chemometric models; analysis of active ingredients in their pharmaceutical dosage forms.