Agriculture Reference
In-Depth Information
2. Application of Multivariate Techniques for the Classification of Soybean Varieties
Cluster analysis was used for searching natural grouping among the studied soybean
lines. This analysis will allow an association of samples based on their similarities in protein
profiles. Hierarchical agglomerative clustering was performed by means of the Ward method
on raw data using squared Euclidean distances as measure of similarity. Three main groups
were observed: two of the clusters consisted of a miscellaneous of soybean varieties from
different continents while the third group (in the middle) was the most homogeneous being
mainly constituted by American varieties.
For the application of discriminant analysis, 64 of the 91 soybean varieties were
employed since they were the ones whose origin data could be checked. 54 of these 64
soybean samples were employed as training samples while the other 10 were used for the
validation of the model. The application of discriminant analysis techniques enabled to obtain
two mathematical functions (DF1 and DF2) that were statistically significant at the 95% of
confidence. These discriminant functions classified the 54 soybean varieties in four
established categories: America, Europe, Asia, and Africa (Figure 4). Soybeans from the
same continent showed certain tendency to present similar chromatographic profiles
observing a good separation among the samples according to their origin. The African
cultivar was located in the corner corresponding to the most negative value of DF1 and the
most positive value of DF2. American and Asian soybean cultivars were differentiated by
DF1 while European cultivars presented intermediate DF2 values. A total of 49 soybean
cultivars from the 54 were correctly classified (percentage of correct classification, 90.7%)
observing the best classification for European and African soybeans while Asian ones were
the worst classified (see Table 2). The most influencing variables for the classification of
soybean cultivars (determined by the stepwise method (forward selection)) were peaks 2 and
5 being possible to obtain a percentage of correct classification of 72.2% using only these two
variables.
Figure 4. Application of discriminant analysis to the area percentages corresponding to 64 different
soybean cultivars from America, Asia, Europe, and Africa.
Search WWH ::
Custom Search