Self organising maps (SOM) represent a particular use of neural networks to perform cluster analysis. The major difference is that instead of a single output a SOM may have hundreds, traditionally arranged in a grid, although any topology can be used. Neighbouring nodes are connected through a neighbourliness parameter which affects the way the weights on the nodes are adjusted as the network is trained. Since each output node is connected to every input node it is possible to find average values for all the instances in a cluster, and as such characterise the cluster.
SOMs are unsupervised, discovering patterns through clusters, and are used in many areas of business including marketing, finance and credit risk analysis.