Learning algorithm for self-organizing map classification of electroencephalogram patterns with individual differences

This paper introduces a new learning algorithm for training self-organizing maps (SOM) to classify electroencephalogram (EEG) patterns that have individual differences. To classify these EEG patterns, we propose an algorithm that specifies the learning area for SOM based on sub-attribute information (SOMSA) related to the individual differences of EEG data. The individual differences are quantified by analysing human personality because we believe that an individual´s personality is responsible for individual differences. In the preprocessing phase, we extract the EEG feature vectors by calculating the time average in each of three frequency bands: θ, α and β. The personality is analysed through ego analysis based on psychological testing. The device for recording EEG is a band-type device with a small number of electrodes. To evaluate the performance of our proposed method, we conducted experiments using real EEG data. Our experimental results show that the accuracy rate of the EEG pattern classification using SOMSA is significantly improved compared with that using standard SOM.