Explainable Emotion Recognition with EEG Wearable Devices

Emotion recognition from electroencephalography (EEG) signals is a key task in affective computing, but the interpretability of machine learning models remains a major challenge, especially in wearable low-density setups. This work presents an explainable pipeline for EEG-based emotion recognition using a 14-channel Emotiv headset under a DEAP-inspired protocol. The framework combines handcrafted spectral and nonlinear features with machine learning classifiers and SHAP-based explanations to analyze the neural factors underlying emotion decoding. Results reveal consistent temporal patterns across subjects, interpretable spatial attribution maps, and a strong contribution of nonlinear complexity descriptors. The findings are consistent with previous analyses on the NeuroSense dataset, suggesting that meaningful EEG markers of emotion recognition remain observable across different low-cost acquisition setups.