Near-infrared spectroscopy for rapid identification of pharmaceutical excipients

This study is based on near-infrared spectroscopic detection technology to achieve 100% classification of in-class and out-of-class pharmaceutical ingredients and excipients by support vector machine model.4 types of 8 different pharmaceutical excipients (starches: corn starch, potato starch, sweet potato starch, pregelatinized starch, maltodextrin, lactose: lactose monohydrate, Cellulose: microcrystalline cellulose, phosphate: magnesium stearate) are collected by near-infrared spectrometer, 150 sets of spectral data each. A total of 1200 spectra are used, 840 spectra of which are randomly divided as the training set and 360 as the validation set. Compare the effects of models built by Bayesian algorithm, support vector machine algorithm, and K-nearest neighbor algorithm paired with first-order difference, second-order difference, MSC, and SNV preprocessing, respectively. The results show that both Bayesian and K-nearest neighbor algorithms achieve 100% out-of-class resolution when paired with first-order difference, MSC, and SG smoothing preprocessing methods, In contrast, the support vector machine achieves 100% classification accuracy without any preprocessing, and the accuracy is not reduced after dimensionality reduction by the competitive adaptive reweighting algorithm. Finally, this experiment achieves 100% accuracy of in-class and out-of-class classification of 8 APIs in 4 classes by NIR spectroscopy combined with support vector machine algorithm model, and the CARS algorithm is used for data dimensionality reduction to simplify the model.