Kohonen created SOM for visualization, not regression or classification.
For classification he created LVQ and LVQ2. All three of Kohonen's algorithms are documented in the NN Toolbox.
For regression and classification the Multilayer Perceptron (NEWFF, NEWFIT and PATTERNNET) and Radial Basis Function (NEWRB) are, typically, superior.
After reading the documentation, try duplicating the demos and examples. Searching for relevant Newsgroup postings may also help.
Write back if you have further questions.
Hope this helps.
Greg
