Unsupervised clustering with growing self-organizing neural network -- a comparison with non-neural approach

Usually used approaches for non-hierarchical clustering of data are well known k-means or k-medoids methods. However, these fundamental methods are poorly applicable in situations where number of clusters is almost unpredictable. Formerly, they were adapted to allow splitting and merging when some defined criterion is met. On the other hand there are also methods based on artificial neural networks concretely on self-organizing maps. One of the interesting ideas in this domain is to allow growing of the net which corresponds to adapted kmeans method. In this article we are going to compare both approaches in a view of ability to detect clusters in unknown data.