Towards a Practical Subject-Independent Affective State Recognition Based on Time-Domain EEG Feature Extraction

This work presents an evaluation of two time domain-based features, i.e., fractal dimension (FD) and higher-order crossings (HOC), for the subject-independent EEG-based recognition of four affective states as a preliminary step towards a practical real-time affective brain computer interface. EEG data were acquired from an experiment targeting the elicitation of four emotions using affective sounds. Features were computed for each electrode individually and tested in terms of classification using the k-nearest neighbors classifier. Results show that the valence of affective states can be recognized effectively, when the arousal level is specified. Moreover, an above chance level classification accuracy was achieved using a single electrode for the four affective states recognition. Both FD and HOC performed similarly, while the best classification rates were achieved from frontal electrode locations.