Cracking the Emotional Code: Identifying Emotions Through Electroencephalograms

In this work, we present a machine-learning model for emotion recognition based on Electroencephalogram (EEG) signals. We employed an online available DREAMERs dataset, from which we classify 4 human emotions such as sadness, joy, anger, and pleasure. In the proposed methodology, first, the acquired EEG signals from each subject were divided into smaller non-overlapping segments of 30 secs. Then, we removed the motion artifact and unwanted noise by utilizing a band-stop (48.5–51.5 Hz) infinite impulse response (IIR) filter, after which we extracted 5 frequency bands. The first 4 frequency bands i.e. Delta (1–4 Hz), Theta (4–8 Hz), Alpha (8–12 Hz), and Beta (12–30 Hz) were then fed to band-pass pass IIR filter, whereas the remaining 5^th frequency band i.e. Gamma (30 Hz) was provided to a high pass filter to extract a total of 6 time and frequency-based features for each band. Adaptive Synthetic (ADASYN) was used to handle imbalance classes. Moreover, Chi-square (CHI2) algorithm was employed to select the top 50 features out of 420 extracted features. Finally, we utilized various classifiers in order to compare the performance of our model. Fine K- nearest neighbor (KNN) classifier demonstrated the highest accuracy of 91.1% with a training time of approximately 25 secs for emotion detection using EEG signals.