The SOM can be used for classification purposes by assigning a class for each reference vector and deciding the class of a sample vector based on the class of its BMU. However, it should be noted that if the class memberships of the training data are known, using the SOM for classification purposes is not sound, since the SOM does not take into account the known class memberships and cannot therefore optimize the class boundaries appropriately. In such cases the Learning Vector Quantizer (LVQ), a close relative of the SOM, or another method of supervised classification should be used .
However, a SOM can be tuned for classification by using a slightly modified learning rule at the end of the training: when updating the reference vectors, if the reference vector has the same class as the input vector, it is moved closer to it, otherwise it is moved away from the input vector (i.e. the sign of the learning coefficient is reversed) . Also some other methods have been used for tuning the SOM for classification: Pham et al. used the SOM to cluster the data, and then teached an MLP with training samples falling to the edges of the clusters to optimize the classification on class boundaries .
It is also possible to use the SOM algorithm for implicit classification of some vector components: the components are left out when searching for the winning unit from the map, but they are updated normally along with the other components. This way the ordering of the map is not influenced by those components and the distribution of their values is based only on their correlation with other vector components.