分析测试学报2025，Vol.44Issue(6)：1131-1138,8.DOI:10.12452/j.fxcsxb.25020465
基于虚拟样本生成的集成模型提升过期药物光谱识别精度
Improving the Accuracy of Spectral Recognition of Expired Drug by an Ensemble Model and Virtual Sample Generation
摘要
Abstract
The qualitative identification of fake drugs based on near-infrared(NIR)spectroscopy needs to extract characteristic information and establish prediction models from complex,overlapped and unstable spectra by using computers and chemometrics.In this kind of task,there may also be an imbalanced classification problem where there are relatively few samples of a certain class.Based on the generation of virtual samples and ensemble modeling,this approach has the potential to im-prove the recognition accuracy for imbalanced training set.In this paper,azithromycin was taken as the research object,a group of experimental samples were designed,and an ensemble algorithm of partial least squares discriminant analysis(PLS-DA)based on virtual samples was proposed to con-struct a classifier for identifying whether a drug sample had expired.The performance of single and ensemble models was compared in ten different spectral ranges,and the influence of different imbal-ance ratios,the composition of minority class samples and ensemble size were also discussed.The sensitivity of ensemble models was improved by about 9%on average.Finally,the overall effective-ness of the ensemble learning strategy was confirmed.The proposed ensemble algorithm shows more advantages when there are too few minority class samples,and the method can also be used for other types of systems.关键词
虚拟样本/集成/近红外光谱/药物/识别Key words
virtual samples/ensemble/near-infrared spectroscopy/drug/identification分类
化学化工引用本文复制引用
谭超,谭成,程斌,邹琴,陈慧,吴同,林瓒..基于虚拟样本生成的集成模型提升过期药物光谱识别精度[J].分析测试学报,2025,44(6):1131-1138,8.基金项目
宜宾学院预研项目(2023YY07) （2023YY07）
过程分析与控制四川省高校重点实验室开放项目(GCFX2023003) （GCFX2023003）
