Emotional Resonance in Brainwaves: EEG based Classification of Emotional Dynamics

This project focuses on combining various machine learning algorithms to classify emotions based on electroencephalogram (EEG) data. In the fields of affective computing, human-computer interface, and healthcare, emotion recognition is significant. The DREAMER (Database for Emotional Analysis in Music Videos) and GAMEEMO datasets, both of which include EEG signals captured during particular stimuli are used in the study. The two datasets are compared at the initial phase of the project in order to figure out which is the most appropriate for additional investigation. The study involves feature extraction, preprocessing, artifact identification, and dataset comparison analysis after dataset selection. Using the selected dataset, several machine learning techniques are used for emotion classification, which include Decision Tree, Random Forest, AdaBoost, Naïve Bayes, and Linear SVM. The results indicate AdaBoost is effective in classifying emotions with the maximum accuracy of 91.7%. Additionally, Adaboost has a F1 score of 94.1% and precision of 88% which tends to be the highest among other algorithms used. Various performance metrics such as F1 Score, Sensitivity, Specificity, Recall and ROC curve are determined for these algorithms, which classify emotions into stress and non-stress classes. Further studies include exploring multimodal approaches and transfer learning to enhance model performance and accurately predict emotions.