Support vector regression and artificial neural network models for stability indicating analysis of mebeverine hydrochloride and sulpiride mixtures in pharmaceutical preparation: a comparative study.
Support vector regression and artificial neural network models for stability indicating analysis of mebeverine hydrochloride and sulpiride mixtures in pharmaceutical preparation: a comparative study.
复制标题
DOI：
10.1016/j.saa.2011.11.003
复制
发表时间：
2012-02
期刊：
影响因子：
--
通讯作者：
Ibrahim A. Naguib;H. Darwish
中科院分区：
文献类型：
--
作者：
Ibrahim A. Naguib;H. Darwish
支持向量回归（SVR）和人工神经网络（ANNs）多元回归方法之间的比较，建立显示每个底层的算法，并进行比较，它们之间的固有的优点和局限性。在本文中，我们比较了支持向量回归机的人工神经网络和变量选择程序（遗传算法（GA））。为了以合理的方式进行比较，通过处理UV光谱数据，将该方法用于在原材料和药物剂型中存在其报告杂质和降解产物（总计6种组分）的情况下，对二元混合物中盐酸美贝维林和舒必利的混合物进行稳定性指示性定量分析，作为案例研究。为了进行适当的分析，建立了6因子5水平实验设计，得到含有不同比例干扰物质的25种混合物的训练集。由5个混合物组成的独立测试集被用来验证所建议的模型的预测能力。所提出的方法（线性支持向量回归机（无GA）和线性GA-ANN）被成功地应用于分析药物片剂含有盐酸美贝维林和舒必利的混合物。结果表明，非线性的问题，以及如何像SVR和ANN模型可以处理它的方法表明，上述多元校正模型的能力，以解卷积的高度重叠的6组分的混合物的紫外光谱，但使用廉价和易于操作的仪器，如紫外分光光度计。
A comparison between support vector regression (SVR) and Artificial Neural Networks (ANNs) multivariate regression methods is established showing the underlying algorithm for each and making a comparison between them to indicate the inherent advantages and limitations. In this paper we compare SVR to ANN with and without variable selection procedure (genetic algorithm (GA)). To project the comparison in a sensible way, the methods are used for the stability indicating quantitative analysis of mixtures of mebeverine hydrochloride and sulpiride in binary mixtures as a case study in presence of their reported impurities and degradation products (summing up to 6 components) in raw materials and pharmaceutical dosage form via handling the UV spectral data. For proper analysis, a 6 factor 5 level experimental design was established resulting in a training set of 25 mixtures containing different ratios of the interfering species. An independent test set consisting of 5 mixtures was used to validate the prediction ability of the suggested models. The proposed methods (linear SVR (without GA) and linear GA-ANN) were successfully applied to the analysis of pharmaceutical tablets containing mebeverine hydrochloride and sulpiride mixtures. The results manifest the problem of nonlinearity and how models like the SVR and ANN can handle it. The methods indicate the ability of the mentioned multivariate calibration models to deconvolute the highly overlapped UV spectra of the 6 components’ mixtures, yet using cheap and easy to handle instruments like the UV spectrophotometer.