A Dempster-Shafer Approach to Multi-Scale EEG Emotion Classification

Brain signal analysis has become a cornerstone in various psychological diagnostics. Electroencephalography (EEG) data, in particular, play a pivotal role in emotion recognition and have recently garnered significant interest within the research community. The primary objective of this study is the classification of human emotional states. To achieve this, a comprehensive computational framework is proposed, encompassing signal preprocessing, feature extraction, feature dimensionality reduction, and classification. The study utilizes the publicly available DEAP dataset, comprising ۳۲-channel EEG recordings from ۳۲ participants. The dataset consists of ۴۰ one-minute trials per participant, sampled at ۵۱۲ Hz, and preprocessed to ensure the removal of physiological artifacts and noise. Feature extraction was performed using Pearson correlation coefficients and Mutual Information (MI) to capture linear and non-linear dependencies, respectively. Subsequently, Principal Component Analysis (PCA) was employed for dimensionality reduction to optimize the feature space. Finally, the study introduces an emotion classification approach based on the Dempster-Shafer (DS) evidence theory. Experimental results demonstrate that the proposed method achieves the highest classification accuracy for the alpha frequency band in the arousal dimension (۵۶.۹۶% ± ۲.۶۰) and for the beta frequency band in the valence dimension (۶۴.۴۱% ± ۱.۱۹). These results, which outperform those obtained from other frequency bands, substantiate the efficacy of the proposed DS-based framework in robustly identifying emotional states.

کلیدواژه ها:

EEG signals ، Dempster-Shafer theory ، Emotion recognition ، Feature fusion