The self-organizing map [Koh95,Bis95] is a popular topology preserving clustering algorithm with nice visualization properties. For simplicity, we only use a 1-dimensional line topology. Also, 2-dimensional or higher dimensional topologies can be used. To generate clusters we use cells in a line topology and train the network for epochs or 10 minutes (whichever comes first). All experiments are run on a dual processor 450 MHz Pentium using the SOM implementation in the Matlab neural network tool-box. The resulting network is subsequently used to generate the label vector from the index of the most activated neuron for each sample. The complexity of this incremental algorithm is and mostly determined by the number of epochs and samples .