Kohonen’s Self Organizing Maps in Excel/VBA, applied for reducing dimensions of colors and of financial ratios from Google Finance

Kohonen’s Self Organizing Maps (SOM) is a type of artificial neural network that is trained using unsupervised learning. The number of dimensions is effectively reduced as a two-dimensional, discrete representation of high-dimensional data is produced. A neighborhood function is used as the topological properties of the input space are preserved.

In this file, the output space is visualized in a 40x40 block of Excel cells that are colored appropriately. If you would like to take a look at a good description of the algorithm, go to AIJunkie’s website.

The second file, chooses three financial indicators – ROE, Debt-to-Equity and Market Capitalization – and once again depicts this three-dimensional data on a two-dimensional SOM. The ratios are from telecom sector of Google Finance and are (0,1) normalized beforehand. As the initial plot is “dominated” by the three biggest companies – for which a substantial portion of the entire space is reserved (the market capitalization ratio is extremely right skewed) – a logarithm transformation is later undertaken. The colors of the graph are unappealing, because the ratio vectors are used as RGB components of Excel cell colors :).