Psychological Stress Classification Based on Coupled Analysis of ECG and EEG Signals

Psychological stress detection methods based on objective physiological signals offer distinct advantages. The bidirectional coupling relationship between the heart and brain has been well-established. We aim to construct a machine learning framework based on heart-brain coupling characteristics to achieve precise stress identification. Utilising EEG and ECG data acquired during resting and mental arithmetic task states. Construct HRV sequences and five sub-band power sequences of EEG. Establish heart-brain coupling combinations. Six coupling features were extracted: cross sample entropy, cross fuzzy entropy, joint distribution entropy, mutual information, magnitudesquared coherence function, and cross power spectral density. Feature selection was performed using recursive feature elimination, followed by classification model construction via support vector machines. The highest classification accuracy (96.67%) was achieved when utilising features from all channels. Furthermore: (1) Cross power spectral density demonstrated exceptional discriminative capability; (2) The non-entropy feature combination exhibited outstanding classification performance; (3) The RR- $\beta$ coupling combination performed exceptionally well. This study confirms the efficacy of heart-brain coupling in identifying psychological states. This finding holds significant value for the early warning and intervention of stress-related disorders.