|
Internet Electronic Journal of Molecular Design - IEJMD, ISSN 1538-6414, CODEN IEJMAT
| ABSTRACT - Internet Electron. J. Mol. Des. May 2003, Volume 2, Number 5, 348-357 |
Support Vector Machines Classification of Black and Green Teas
Based on Their Metal Content
Ovidiu Ivanciuc
Internet Electron. J. Mol. Des. 2003, 2, 348-357
|
Abstract:
Green and black teas are made from the processed leaves of
Camellia sinensis. The metal content (Zn, Mn, Mg, Cu, Al, Ca,
Ba, and K) of commercial tea samples, determined by
inductively coupled plasma atomic emission spectroscopy, can
be used in pattern recognition models to discriminate between
the two tea types. We have investigated the application of SVM
(support vector machines) for the classification of 44 tea samples
(26 black tea and 18 green tea) based on the metal content. An
efficient algorithm was tested for the selection of input
parameters for the SVM models, in order to find the minimum
metal profile that provides a good separation of the two classes.
Using the hierarchical descriptor selection procedure, the initial
group of eight metals was reduced to a set of three metals,
namely Al, Ba, and K. The classification of the green and black
teas was done with the dot, polynomial, radial basis function,
neural, and anova kernels. The calibration and leave-20%-out
cross-validation results show that the statistical performances of
SVM models depend strongly on input descriptors, kernel type
and various parameters that control the kernel shape. Several
SVM models obtained with the anova kernel offered the best
results, all with no error in calibration and one error in prediction
(for a green tea sample). The hierarchical descriptor selection
algorithm is an effective procedure to identify the optimum set of
input variables for an SVM model. Using the Al, Ba, and K
content determined with the inductively coupled plasma atomic
emission spectroscopy, a highly predictive SVM model was
developed for the classification of green and black teas.
|