Three Dimensional Emotion State Classification based on EEG via Empirical Mode Decomposition

Electroencephalography (EEG) is useful for mapping emotions directly from the brain, but its heterogeneous signals make it challenging to extract features accurately. Prior works for emotion classification uses EEG data without removing data heterogeneity leading to misclassification or inaccurate classification. This paper proposes an EMD-based methodology for EEG data that segments signals into multiple IMFs to remove heterogeneity and extract significant features. The proposed approach uses a Feed-Forward Neural Network (FFNN) to classify emotions via the VAD model and shows a 5-6% increment in accuracy, precision, and recall scores for emotion classification. Experimental results demonstrate good evaluation performance scores for classifying emotional states on two publicly accessible emotional datasets, AMIGOS and DREAMER.