EEG-based Emotion Recognition: Assessing Misclassification Rates Using Machine Learning Techniques

Abstract

Everyday human interaction heavily relies on emotions, which play a pivotal role in fostering genuine human-machine interaction. Emotional equilibrium is crucial for individual well-being, with regular meditation being a widely acknowledged means to achieve it. This study delves into the effect of meditation on emotional responses using Electroencephalography (EEG) technology. EEG can establish correlations between mental attributes and emotional states, with four emotions (Thrilled, Angry, Sad, and Relax) used as categorized stimuli based on valence-stimulation. Functional connectivity in EEG brainwave activity is compared during these four emotions both before and after meditation sessions. Results demonstrate that meditation promotes more cohesive emotional experiences, albeit with a reduction in classification accuracy observed after an 8-week meditation regimen. This research utilizes EEG data to develop a classification system for discerning familiarity levels within EEG signals, employing the Hjorth Descriptor to condense signal characteristics into three distinct criteria. A Multilayer Perceptron classifier, leveraging input parameters, achieves a peak accuracy of 96% through a combined application of three functions.