Using Selforganizing Feature Maps to Classify EEG Coherence Maps

In this work we have been applying self-organizing feature maps [3] to the problem of unsupervised classification of EEG data. The type of EEG used are so-called coherence maps based on 19 electrodes, which were derived during specific cognitive taks such as mental rotation. The goal was to exploit the network learning scheme as extractor for any task- (or other parameter-) related information in the data. In other words, we used the self-organizing feature maps to detect whether the EEG inputs can be classified acording to underlying parameters such as the type of task performed. This paper reports about the very promising results of the experiments.