Partial least squares and linear support vector regression chemometric models for analysis of Norfloxacin and...

The aim of the presented work is to compare two popular chemometric methods which are partial least squares regression (PLSR) and support vector regression (SVR). The comparison shows their characteristics via application of the suggested methods to analysis of Norfloxacin (NF) and Tinidazole (TZ) with the presence of a potential impurity of Tinidazole; 2-Methyl-5-nitro-1H-imidazole (MNZ). For appropriate analysis, a 3 factor 4 level experimental design was constructed, which results in a training set composed of 16 mixtures which contains different concentrations of the three components; achieving symmetry, rotatability and orthogonality in mixture space. In order to validate the prediction ability of the suggested models, an independent test set consisting of 8 in-space and 8 out-of-space mixtures was used. The presented results show high specificity and accuracy of the mentioned multivariate calibration models for analysis of in-space samples of NF and TZ in presence of (MNZ) using UV spectral data. Statistical comparisons of predictive abilities of proposed models against classical least squares CLS model and against each other was performed; whether for analysis of test set mixtures or dosage form. CLS model showed lower predictive ability compared to other models. Results obtained by SVR model are as accurate as PLSR model, however, optimization and implementation of PLSR is faster and easier, hence PLSR could be of choice for this given case study. The developed chemometric models were validated as directed by ICH strategies. The validated methods were efficiently used for estimation of NF and TZ in pure powders and pharmaceuticals which indicates their suitability for application in quality control examination of both of the drugs.