EmoClass: Subject-Independent Emotion Classification for BCI Systems Using EEG Signals

Human-Computer Interaction (HCI) explores the interaction between humans and technology, considering emotional responses as an integral part of human life. Emotions are significant influencers of decision-making, and public well-being, intricately linked to mental health, detectable through BrainComputer Interface (BCI) technologies. BCI provides insights into neural correlates of emotional states and aids in mental health assessment and intervention strategies. Electroencephalography (EEG) based BCI systems play a significant role in classifying emotions to enhance human interaction. The proposed models were applied to a PhyMER dataset for emotion classification, encompassing both binary and multi-class models using a subject-independent approach. The features are extracted from EEG signals using Fast Fourier Transform (FFT) and Peak-to-Peak (P2P) analysis. The proposed binary and multiclass models were analyzed to classify the emotions using a variety of machine learning (ML) and deep learning (DL) algorithms. The performance metrics of the binary and multiclass ML models have maximum F1 scores of 62.04 and 28.48 respectively. Additionally, the binary and multiclass DL model achieved a maximum F1 score of 61.31 and 29.2 respectively.