There is a large amount of analytical methods for analyzing data, from classical statistical approaches such as hypothesis tests and linear regression to the most complicated machine learning methods, like Artificial Neural Networks, Random Forest or Bagging and other ensemble methods. Besides, there are the so-called visual techniques, which can provide a visual intuition about data.
Visual Techniques can be mainly divided into 2 groups: univariate and multivariate. The univariate representations produce representations of one variable while the multivarivate representations try to show the relationship between several variables. Some examples of univariate representations are: histograms, box-and-whiskers plots, etc. Within multivariate representations, Parallel Coordinates and Self Organizing Maps (a.k.a. Kohonen Maps) must be highlighted as most representative elements of this kind of techniques. In this post, I’m going to try to explain how Self Organizing Maps (SOM hereafter) work and how to interpret it so that you’re going to see how powerful are them. I want to explain how they are built but if you want to avoid the mathematical part, you can go below and see the example and how to interpret it.
How Self Organizing Maps work
The SOM was proposed in 1984 by Teuvo Kohonen, a Finnish academician. It is based in the process of task clustering that occurs in our brain; it is a kind of neural network used for the visualization of high-dimensional data. It compresses the information of high-dimensional data into geometric relationships onto a low-dimensional representation. In a SOM algorithm, the neurons are ordered in two layers: the input layer and the competition layer. The input layer is composed of N neurons, one for each input variable. The competition layer is composed of a topological low-dimensional grid of neurons (usually 2-dimensional) geometrically ordered. The number of neurons in the output layer depends on the problem. It may fluctuate from a few to several thousands depending on the complexity of the problem.