FREQ-EER: A Novel Frequency-Driven Ensemble Framework for Emotion Recognition and Classification of EEG Signals

Emotion recognition using electroencephalogram (EEG) signals has gained significant attention due to its potential applications in human–computer interaction (HCI), brain computer interfaces (BCIs), mental health monitoring, etc. Although deep learning (DL) techniques have shown impressive performance in this domain, they often require large datasets and high computational resources and offer limited interpretability, limiting their practical deployment. To address these issues, this paper presents a novel frequency-driven ensemble framework for electroencephalogram-based emotion recognition (FREQ-EER), an ensemble of lightweight machine learning (ML) classifiers with a frequency-based data augmentation strategy tailored for effective emotion recognition in low-data EEG scenarios. Our work focuses on the targeted analysis of specific EEG frequency bands and brain regions, enabling a deeper understanding of how distinct neural components contribute to the emotional states. To validate the robustness of the proposed FREQ-EER, the widely recognized DEAP (database for emotion analysis using physiological signals) dataset, SEED (SJTU emotion EEG dataset), and GAMEEMO (database for an emotion recognition system based on EEG signals and various computer games) were considered for the experiment. On the DEAP dataset, classification accuracies of up to 96% for specific emotion classes were achieved, while on the SEED and GAMEEMO, it maintained 97.04% and 98.6% overall accuracies, respectively, with nearly perfect AUC values confirming the frameworks efficiency, interpretability, and generalizability.