EEG-based automatic emotion recognition: Feature extraction, selection and classification methods

摘要

Automatic emotion recognition is an interdisciplinary research field which deals with the algorithmic detection of human affect, e.g. anger or sadness, from a variety of sources, such as speech or facial gestures. Apart from the obvious usage for industry applications in human-robot interaction, acquiring the emotional state of a person automatically also is of great potential for the health domain, especially in psychology and psychiatry. Here, evaluation of human emotion is often done using oral feedback or questionnaires during doctor-patient sessions. However, this can be perceived as intrusive by the patient. Furthermore, the evaluation can only be done in a noncontinuous manner, e.g. once a week during therapy sessions. In contrast, using automatic emotion detection, the affect state of a person can be evaluated in a continuous non-intrusive manner, for example to detect early on-sets of depression. An additional benefit of automatic emotion recognition is the objectivity of such an approach, which is not influenced by the perception of the patient and the doctor. To reach the goal of objectivity, it is important, that the source of the emotion is not easily manipulable, e.g. as in the speech modality. To circumvent this caveat, novel approaches in emotion detection research the potential of using physiological measures, such as galvanic skin sensors or pulse meters. In this paper we outline a way of detecting emotion from brain waves, i.e., EEG data. While EEG allows for a continuous, real-time automatic emotion recognition, it furthermore has the charm of measuring the affect close to the point of emergence: the brain. Using EEG data for emotion detection is nevertheless a challenging task: Which features, EEG channel locations and frequency bands are best suited for is an issue of ongoing research. In this paper we evaluate the use of state of the art feature extraction, feature selection and classification algorithms for EEG emotion classification using data from the de facto standard dataset, DEAP. Moreover, we present results that help choose methods to enhance classification performance while simultaneously reducing computational complexity.