Master Thesis Seminar: "Modelling and Analysis of High Order Chemical Data"
Date: February 18, 2005 (Friday) at 10:15
Analytical chemistry is a science with strive is to determine the amounts of entities in a chemical sample. The existing sampling techniques deliver abundance of data about the di erent molecules characteristics. To extract information out of this multivariate data delivered is the main occupation of the analytical researchers. Analyses of nuclear magnetic resonance spectroscopy
data are today done with the decomposing methods of PCA and PARAFAC. As an attempt to find complementing information in the data, a non-linear pattern recognition method with neural network is developed. A selforganizing map (SOM) algorithm with its result projection on a Kohonen topmap, is implemented and
analysed in comparison with PCA and PARAFAC. To compare with PARAFAC the algorithm has to be able to handle three-way input as well as two-way. The SOM
is investigated by its neurons weights out of di erent clusters, and examination of the topmap. Comparison is done between the variables indices and loadings within the SOM and between the three methods. Results from the SOM shows that it is classifying the samples, not the same but similarly with PCA and PARAFAC. The sorting procedure emphasizes almost same variables as PCA and
projects the result in a alternative way. The three-way data need more tests but has potentials for being a good complement to PARAFAC.
Last modified Dec 9, 2011 12:59 pm