For these experiments a Hidden Markov Modell (HMM) were used. 
A flat-start training was relized using the Baum-Welch algorithm 
,decoding was made with the Viterbi algorithm. HMMs with three - five 
states were used, depending on the particular training set. The 
Training was accomplished with data from five electroed (C5, C3, Cz, C2, C4). 
90 samples were taken together for one trainings-vector.Hence, 0.9 sec.
from trainingsdata were observed and classified in one of the three
categories. No preprocessing was made.