Inter-Subject Emotion Detection Using Empirical Mode Decomposition on EEG Data

Accurate identification of human emotional states can be achieved through various methods, including subjective self-reports and obj ective measurements like facial expressions, vocal patterns, and physiological signals. Among these, use of physiological signal such as electroencephalogram (EEG) is optimal for emotion detection due to its direct measurement of brain activity, providing insights into neural processes underpinning emotions. However, detecting emotions using EEG signals is challenging in subj ect-independent scenarios, due to the fluctuating and unpredictable nature of these brain wave recordings. This study addresses these challenges by utilizing empirical mode decomposition (EMD) to analyze EEG signals. EMD is an iterative method that decomposes signals into intrinsic mode functions (IMFs) without needing predefined basic functions. A 4-second sliding window with a 2-second overlap is utilized to divide the EEG signal into segments. Each segment is decomposed into five IMFs, and nonlinear features such as Shannon entropy, differential entropy, collision entropy, Hjorth parameters, and Higuchi's fractal dimension are extracted. These features serve as 2D inputs to an EEGN et classifier, a convolutional neural network designed for EEG signals. The study employs the DEAP dataset, which includes 32-channel EEG recordings from 32 participants who rated their arousal and valence levels after viewing music video excerpts. These ratings are mapped to high versus low arousal and valence categories. The study focuses on inter-subj ect emotion recognition and the findings from our experiments showcase average classification accuracies of 65.46% (±0.09 (SD)) for arousal and 62.5% (±0.06 (SD)) for valence in inter-subject emotion recognition.