Using Machine Learning to Model EEG-Derived Brain Activity During Emotion Regulation

Emotion Regulation (ER) is the ability to manage emotional responses. ER is important for maintaining mental health and handling social interactions, especially under stress. This study explores the brain activity involved in ER using electroencephalography (EEG) and machine learning (ML) models to predict successful and unsuccessful ER. Study participants viewed emotional and neutral images under two conditions: regular viewing and being asked to reduce their emotions. At the end of each experimental trial, participants rated the intensity of their emotional response to the image. Ratings of low intensity (1 and 2) were classified as successful ER, whereas ratings of high intensity (3 and 4) were considered indicative of unsuccessful ER. EEG signals were analyzed in both time and frequency domains to identify patterns linked to ER. In the time domain, significant differences in Global Field Power (GFP) were observed, especially in the frontal and central regions of the brain. Frequencydomain analysis using Power Spectral Density (PSD) showed that theta, beta, and gamma bands were important for regulating emotions. Using these analysis results, machine learning models were trained to predict regulation success. Among the models, a neural network with Maximum Mean Discrepancy (MMD) loss performed the best, achieving an F1-score macro of 75.57% with a subject-independent approach. These machine-learning models highlight the importance of frontal and central brain regions and beta brain frequency signals in the prediction of ER levels. It shows that combining EEG data with advanced machine learning methods can create accurate models for understanding and predicting emotional responses. Additionally, this integrated EEG-based approach represents a novel framework for ER assessment, offering a promising direction for future research and enabling personalized mental health treatments.