Time-frequency analysis of EEG for improved classification of emotion

Abstract

Emotion detection has crucial role in many domains especially in health and e-learning sector. This study aims to improve the accuracy in detecting emotions using brain activity. It addresses two primary problems associated with current emotion recognition systems. Firstly, these existing systems can classify only small classes of emotion. Secondly, analysis of the EEG is complex due to its non-stationary and non-linear characteristics. We conducted experiments to record EEG of subjects using 14 electrodes attached directly to the scalp based on International 1020 system. To remove artefacts, raw signals are pre-processed. Emotional patterns associated with EEG are detected on time-frequency domain using Hilbert–Huang Transform (HHT). Multiclass Support Vector Machine classifier (MC-SVM) is used to distinguish emotions from recorded data based on the instantaneous frequency obtained through HHT. The results revealed the effectiveness of the suggested time-frequency-based analysis method to detect wide range of emotions using EEG signals.