Human Emotion Recognition with a Regression Classifier Based on a New Feature Definition for Multi-channel EEG Waveforms

A new feature definition for multi-channel EEG waveforms, which involves estimated second-order statistics, e.g. autocorrelation, of individual channel waveforms, is presented. Based on new features, a vector description with randomly chosen length and time lags as a multivariate process is employed to model human emotions. A regression classifier is designed and simulated to estimate a set of human emotion states based on the derived feature vector. Experiments with a real-world publicly available dataset indicate that the new feature and associated vector descriptions with chosen classifier lead to successful recognition of human emotion states. The simplicity and straightforward modeling of human emotions with use of new approach is expected to lead improved-performance human-computer interface systems in real-time in predictive manner.