ICP-AES结合主成分分析和决策树模型的四种品牌白酒鉴别方法研究
Study on discrimination of four Chinese brand spirits based on ICP-AES coupled the principal component and decision tree analysis
-
摘要: 采用电感耦合等离子体原子发射光谱（ICP-AES）测定了四种品牌56个白酒样品（五粮液,郎酒,全兴,五津醇）中的16种元素含量。通过对结果进行z-score标准化,消除各元素间量纲差异,再对其进行主成分分析。结果表明,第一主成分的方差贡献率为40.3%,前十主成分的贡献率达96.3%,基本保留了原变量的所有信息。选择前十主成分建立决策树分类预测模型,模型的交叉验证准确率高达97.6%,再用模型预测未参与建模的15个白酒样品,准确率高达100%。模型能够准确区分五粮液,郎酒,全兴,五津醇四种品牌白酒。Abstract: The potential of ICP- AES for metal element profiling of Chinese spirit samples was examined. Sixteen elements in fifty six spirits samples representing four varieties of brands( Wuliangye,Lang Liquor,Quanxing,Wujinchun) were determined.The set of data was employed to construct a sample class prediction model based on z- score standardization followed by principal component analysis( PCA) and decision tree analysis( DT),which was employed to explore the structure of the data and construct classification and prediction model. The first principal component explained 40.3% of variance while the top ten components explained 96.3% of variance which was employed to construct the DT model. The validated DT model based on 5- fold cross- validation enabled correct classification of 97.6% of samples,and other 15 spirit samples could be predict correctly. The Wuliangye,Lang Liquor,Quanxing,Wujinchun could be classified intensively.
下载: