Decoding emotional patterns using NIG modeling of EEG signals in the CEEMDAN domain

Electroencephalogram (EEG) signals used for emotion classification are vital in the Human–Computer Interface (HCI), which has gained a lot of focus. However, the irregular and non-stationary characteristics of the EEG signals manifest barriers and limit state-of-the-art techniques from accurately assessing different emotions from the EEG data, leading to minimal emotion recognition performance. Moreover, cross-subject emotion recognition (CSER) has always been challenging due to the weak generality of features from EEG signals among subjects. Thus, this study employed a novel algorithm, Complete Ensemble Empirical Mode Decomposition with adaptive noise (CEEMDAN), which decomposes EEG into intrinsic mode functions (IMFs) to comprehend the associated EEG’s stochastic characteristics. Further IMFs are characterized by Normal Inverse Gaussian (NIG) probability density function (PDF) parameters. These NIG features are fed into an optimized Extreme Gradient Boosting (XGboost) classifier developed using a cross-validation technique. The uniqueness of this research is in the use of NIG modeling of CEEMDAN domain IMFs to extract specific emotions from EEG signals. Qualitative, visual, and statistical assessments are used to illustrate the importance of the NIG parameters. Extensive experiments are carried out with the online available data sources SJTU Emotion EEG Dataset (SEED), SEED-IV, and Database for Emotion Analysis of Physiological Signals (DEAP) to evaluate the potency of the proposed approach. The suggested system for recognizing emotions performed better than cutting-edge techniques, attaining the highest accuracy of 98.9%, 97.8%, and 96.7% with the tenfold cross-validation (CV) protocol and 96.84%, 95.38%, and 91.39% for cross-subject validation (CSV) approach using SEED, SEED-IV, and DEAP databases, respectively.