基于Stacking集成学习的马兜铃酸及其类似物鉴别

Discrimination of aristolochic acid and its analogues based on stacking ensemble learning

以中草药中所含成分马兜铃酸及其类似物为研究对象,针对传统中药鉴定存在的主观性强、操作复杂等不足以及单一机器学习模型鉴别精度不高的问题,提出多模型融合的Stacking集成学习分类模型,用来实现马兜铃酸及其类似物的鉴别。采集马兜铃酸、1,10-菲咯啉-4,7-二甲酸、菲醌、β-谷甾醇4种样品的近红外光谱数据,对其进行数据预处理与主成分分析降维,基于降维后的数据特征,通过遍历搜索策略构建了以随机森林、支持向量机、朴素贝叶斯为基分类器,随机森林为元分类器的Stacking集成学习分类模型。结果表明,Stacking集成学习分类模型具有最佳表现性能,鉴别正确率最高达到99.38%,比K最近邻、决策树、随机森林、支持向量机、朴素贝叶斯分类模型的平均鉴别正确率高8.23个百分点,并且在精确率、召回率、综合评价指标(F1值)方面有优异表现。综上可见,本研究提出的Stacking集成学习分类模型能够快速有效地鉴别马兜铃酸及其类似物。

Aristolochic acid and its analogues contained in Chinese herbal medicine were taken as the research objects.Classification model based on Stacking ensemble-learning with multi-model fusion was proposed to identify aristolochic acid and its analogues,aiming at the shortcomings in traditional Chinese medicine identification such as strong subjectivity,complex operations and low accuracy of single classifier model.The near-infrared spectroscopy data of aristolochic acid,1,10-phenanthroline-4,7-dicarboxylic acid,phenanthraquinone andβ-sitosterol samples were collected.The data were preprocessed and principal component analysis was used to reduce dimensionality.Stacking ensemble-learning model was constructed through traversal search strategies based on the data features after dimensionality reduction,with random forest(RF),support vector machine(SVM),naive bayes(NB)as base classifiers and RF as meta classifier.The results showed that classification model based on Stacking ensemble-learning showed the best performance,with a discrimination accuracy rate of 99.38%,which was 8.23 percentage point higher than the average discrimination accuracy rate of classification models like K nearest,decision tree,RF,SVM and NB.Moreover,the proposed method showed excellent performance in precision,recall ratio and comprehensive evaluation index(F1 score).Therefore,the method proposed in this study can quickly and effectively identify aristolochic acid and its analogues.