Brain Neural Data Analysis with Feature Space Defined by Descriptive Statistics

摘要

We consider learning to discriminate emotional states of human subjects, based on their brain activity observed via Electroencephalogram (EEG). EEG signals are collected while subjects were viewing high arousal images with positive or negative emotional content. This problem is important because such classifiers constitute "virtual sensors" of hidden emotional states, which are useful in psychology science research and clinical applications. The feature selection has a major role. Recently we have proposed a sequential feature selection (SFS) procedure that reduced the inherent data variability among subjects and led to a high inter-subject emotion recognition accuracy (98 %). However the SFS is a computationally intensive approach that is difficult to apply to any classification model. In this paper we extend that line of research and propose a computationally less involved feature selection technique based on descriptive statistics (mean and standard deviation) of the neural signatures across subjects. This approach reveals to be a good compromise between prediction accuracy and numerical complexity.