Multi-Channel EEG-based Multi-Class Emotion Recognition From Multiple Frequency Bands

Electroencephalogram (EEG)-based emotion recognition has demonstrated encouraging results using machine learning (ML)-based algorithms. This study compares the performance of different frequency bands using four ML-based classifiers for the recognition of multi-class human emotions from EEG signals. Initially, the raw EEG signals are divided into five frequency bands such as delta, theta, alpha, beta, and gamma bands. Secondly, the statistical, time and frequency domain features are extracted. To classify emotions into positive, negative and neutral classes from the SEED dataset, these features are fed to four ML-based classifiers. This study shows the efficacy of an ensemble ML-based classifier over traditional classifiers. The best highest average classification accuracy reported by the random forest (RF) classifier for the delta band is 95.71%. The second highest average accuracy was reported by KNN with 80.32% for the theta band. A similar trend was also followed by other frequency bands. In conclusion, our study demonstrated the value of the proposed ML-based model for multi-class emotion recognition.