Performance evaluation of classifiers in distinguishing mental tasks from EEG signals

During the last decade a lot of research has been done to study the possibility of voluntarily controlling machines by EEG signals. To accomplish that, very accurate recognition of the intended task is needed and so, given the variability of the human brain’s signals, discriminant features and high performance classifiers are demanded. In this paper, a study on the performance of different classifiers for distinguishing three mental tasks using EEG signals, is presented. The acquired EEG data is filtered and processed, and a set of nine features about power, the signals’ synchronization, and instantaneous frequency are extracted. Classification performance was analyzed across subjects using seventeen individual classifiers. The results obtained from each classifier were also combined in order to evaluate both the efficiency of individual classifiers and the use of combiners.