Determination of terazosin in the presence of prazosin: Different state-of-the-art machine learning algorithms with UV spectroscopy.

Abstract

Counterfeit drugs have adverse effects on public health; chromatographic methods can be used but they are costly. In this study, we developed cost-effective and environmentally friendly methodology for the analysis of terazosin HCl (TZ) in the presence prazosin hydrochloride (PZ) using UV spectroscopy in conjunction with machine learning (ML) models. Variable selection algorithms were applied to select most informative spectral variables. Thirty-five ML models were assessed and their performances were compared. The models covered a wide range of prediction mechanisms, such as tree-based, linear, self-organizing maps, neural network, Gaussian process, boosting, bagging, Bayesian models, kernel methods, and quantile regression. The values of the root mean square error (RMSE), coefficient of determination (R2), and absolute mean error (MAE) were obtained for the evaluation of the developed models. According to the results of these performance indices, linear model showed the highest prediction capacity among all other models. RMSE, R2 and MAE values of (0.159, 0.997 and 0.131) and (0.196, 0.99 and 0.161) were obtained for train and test datasets, respectively. The predictive models in this study can be useful for the researchers who are interested to work on the determination of active ingredients in pharmaceutical dosage forms in the presence of interference using UV spectroscopy; therefore, it was used to determine TZ without interference of PZ.