0

Can anyone tell me if it's possible and how to create neural network from a matrix 60 x 5, which describes 60 cities in 5 different parameters (population, bugdet, area, schools, hospitals). The final results should be presented as following: similar (cities)nodes should be grouped close - I want to see diversity and similarity of collected data. Any help and suggestion is welcome. best regards

user3237786
  • 11
  • 2
  • Following a machine learning course might be useful. But this won't work, neural networks are for regression and classification, not clustering. – Noctua Jan 26 '14 at 15:44
  • @Noctua: of course and you can use neural networks for clustering, SOM for example, check the bibliography for more. – pater Jan 27 '14 at 14:15
  • The number of nodes should be smaller. Although you can extract the clusters from the similarity of the nodes, why don't let some very similar cities to be grouped at the same node. How smaller it should be, it depends on the analysis of the results that you desire. If you want to create large groups of cities for some reason then choose a small number. If want to explore the data choose a bigger number. In any case your numbers are too small so you can easily do both :) – pater Jan 27 '14 at 14:20
  • I'm not sure what you are asking: the SOM is a reasonable way to do exactly what you want; if you're using matlab, you should be able to just use the SOMToolkit (maybe it's in machine learning now) out of the box. If you don't have access to it, you should look into the research literature. – Dave Jan 27 '14 at 17:31
  • See also: (http://stackoverflow.com/questions/13074930/identify-clusters-in-som-self-organizing-map) – Dave Jan 27 '14 at 17:32

1 Answers1

1

You are referring to a clustering result.

This could be acheived using a Self-Organizing Map (SOM) or any other variation of SOM. I would suggest you to go through the wiki link I have given to know more about the SOM.

If you choose to use the SOM and has one set up try experimenting with the map size and the learning rate to find the optimal level of clustering required.

You can use the following links to find implementations of SOM

  1. Matlab
  2. Java and more variations of SOM
Synex
  • 207
  • 6
  • 18