Characterisation of cocoa butters and other vegetable fats by pyrolysis-mass spectrometry
Pyrolysis mass spectrometry (Py-MS) provides a fast way for the characterization of samples and does not require tedious sample preparation. However, the mass spectra obtained by this method are not interpretable by itself and multivariate data analysis in the meaning of pattern analysis is required. Here Py-MS was used for the discrimination of cocoa butters from other vegetable fats. Mass spectra ranging from 50 amu to 250 amu were analyzed by principal component analysis (PCA) and with neural nets. The supervised learning of neural nets leads to a good discrimination between the two classes. Detailed analysis of the nets revealed that only the first 60 masses were used within the net. The use of PCA, as an unsupervised pattern recognition technique, requires a careful selection of the number of masses included in the calculation. Canonical variance analysis was applied to determine the significant masses. Optimal performance of PCA was observed only using the first 22 significant masses. Most of these masses were different from the ones used by the neural net. It seems that the mass spectra obtained by Py-MS contain sufficient information for the discrimination of pure cocoa butter from other vegetable fats, but none of the methods seems to be able to extract all information available. Neural net provides a very robust method for this task and no prior data selection was necessary.
Bibliographic Reference: Article: Fresenius Journal of Analytical Chemistry
Record Number: 199611165 / Last updated on: 1996-10-28
Original language: en
Available languages: en