کد مقاله | کد نشریه | سال انتشار | مقاله انگلیسی | نسخه تمام متن |
---|---|---|---|---|
1179784 | 1491542 | 2014 | 6 صفحه PDF | دانلود رایگان |
• The concentrations of 27 elements in 128 M. tenacissima samples were analyzed.
• Three machine learning tools were applied to identify the origin.
• Pattern recognition procedures displayed different degrees of success.
• SVM-grid model is superior to all the other mathematical methods.
Multi-element analysis of Marsdenia tenacissima samples was used to develop a reliable method of tracing the geographical origins. 27 elements in 128 samples from four provinces of China were analyzed by inductively coupled plasma-atomic emission spectroscopy. First, we used ANOVA and PCA for a preliminary analysis. Then, machine-learning tools for identification and classification were applied to verify the feasibility of using data mining tools to find the region where M. tenacissima samples originated. The study clearly indicated that the support vector machine, the radial basis function neural network, and the random forest chemometric tools had the potential to identify the origins of M. tenacissima. The results revealed that the SVM-grid model was superior to all the other mathematical methods with an average discrimination rate of 98.875% for the training set and 100% for the test set. The order of successful identification rates is as follows: SVM-grid > SVM-ga > RF > RBF-NN > SVM-pso. Moreover, there have been few relevant studies about the application of these machine-learning tools combined with multi-element analysis for tracing the geographical source, so this paper can serve as a reference to identify the origin and perform quality assurance in the field of medicinal plants.
Journal: Chemometrics and Intelligent Laboratory Systems - Volume 136, 15 August 2014, Pages 115–120