Application of Empirical Mode Decomposition for Feature Extraction from EEG Signals

Performance of any brain computer interface system depends upon features of electroencephalogram signals. Electroencephalogram signals undergo for unpredictable changes when vigilance state of human brain alters widely. This may cause adverse changes in extracted features and affect classification performance of brain computer interface system. To avoid miss-classification, brain computer interface should obtain alertness level of user periodically. The aim of present work is to analyze effectiveness of empirical mode decomposition based fractal feature extraction methodology of electroencephalogram signals, for the identification of the two different mental conditions i.e. alert and drowsy. Proposed methodology of feature extraction is occurred in three steps. In the first step, two types of electroencephalogram signals (i.e. alert and drowsy) are acquired from six healthy subjects and decomposed into sub-bands using empirical mode decomposition technique. Significant instantaneous frequency vectors are calculated from decomposed coefficients in the second step. In the third step, two fractal dimensions are computed from instantaneous frequency vectors, as two independent feature vectors of electroencephalogram signals. The prepared feature vectors are used as an input to support vector machine, artificial neural network and random forest tree classifier for classification.