Self-Organizing Maps

A Self Organizing Map is a powerful tool for visualization and data clustering. The basic principle of SOM is that it converts data with a high dimensionality into data with a low dimension. This does not mean that the data loses initial patterns of the data; these patterns are still present in the lower dimension. It means that it uses the first law of geography: things that are near to each other are more related than things that are further away from each other (Jiang & Harrie, 2004).

Maybe the most famous example of a Self-Organizing Map is the world poverty map (figure 4). This map is based on statistics provided by the World Bank for the year 1992. In total there were 39 variables, that where used to indicate poverty. Examples of these variables are: education, health, nutrition etc. Because of the high amount of variables, SOM is a perfect solution to map these variables. The legend shows the distribution of the poverty (figure 5). In the lower right corner, countries with the highest poverty were displayed, while the counties with the least poverty were displayed in the upper left corner (Neural Networks Research Centre, 1997).

 

Figure 4. World Poverty map, created with SOM (Neural Networks Research Centre, 1997).



Figure 5. World Poverty map legend. Countries with highest poverty in the lower right, countries with the least poverty in the upper left (Neural Networks Research Centre, 1997).


Jiang, B., & Harrie, L. (2004). Selection of Streets from a Network Using Self-Organizing Maps. (Oxford, Ed.) Transactions in GIS, 8 (3), 335-350.

Neural Networks Research Centre, Helsinki University of Technology. (1997). World Poverty Map. http://www.cis.hut.fi: http://www.cis.hut.fi/research/som-research/worldmap.html, r etrieved Oktober 27, 2015.