FLSOM with different rates for classification in imbalanced datasets

Abstract

There are several successful approaches dealing with imbalanced datasets. In this paper, the Fuzzy Labeled Self-Organizing Map (FLSOM) is extended to work with that type of data. The proposed approach is based on assigning two different values in the learning rate depending on the data vector membership of the class. The technique is tested with several datasets and compared with other approaches. The results seem to prove that FLSOM with different rates is a suitable tool and allows understanding and visualizing the data such as overlapped clusters. © Springer-Verlag Berlin Heidelberg 2008.