Multifaceted Discrete Emotion Recognition from EEG Physiological Signals via Machine Learning Techniques

Emotions, intricate mental states shaped by neu-rophysiological alterations influenced by cognitive processes, sensory inputs, behaviours, and diverse experiences, remain a compelling domain for exploration. The Electroencephalogram (EEG) is a potent tool that directly quantifies these emotional variations by capturing brain signals. Recent strides in emotion recognition have harnessed traditional machine learning classifiers to automate the identification of human emotions with remarkable success. This paper delves into the relatively underexplored Discrete Emotion Model, unveiling its capacity to achieve outstanding accuracy despite historical reservations regarding its effectiveness, using the ECSMP (Emotion, Cognition, Sleep, and Multi-model Physiological signals) dataset. It has two distinct environments, Video watching, and CANTAB-based cognitive assessment phases, enabling seamless data collection and analysis. This research effectively quantifies emotion for binary classification (classifies the type of emotion felt) and multiclass classification (classifies the intensity of emotion felt), elevating emotion recognition capabilities through the synergy of EEG technology. The exceptional performance of XGBoost, with a 96.5% accuracy rate in binary classification and 95% in multiclass emotion recognition, highlights its prowess compared to the other models tested.