Publication

Abstract :

Emotion recognition using physiological signals has shown a lot of attention because of its major applications in healthcare and human computer interaction. This study looks into a multimodal approach for emotion recognition using EEG, ECG, EMG, and fNIRS signals for efficient emotion classification. A preprocessing pipeline to remove noise, artifacts, and flat signals was implemented, followed by segmentation using sliding windows. Both linear (traditional statistical measures like mean, variance, standard deviation and root mean square) and nonlinear features (Sample Entropy, Permutation Entropy, and Hjorth parameters) were extracted to get the emotional cues within physiological responses. Experimental results demonstrate that EEG features, mainly linear features and combination of linear and non-linear features combined with machine learning classifiers like LabelPropagation and ExtraTrees, give high ac- curacies, with an accuracy of 1.00. In contrast, unimodal ECG, EMG, and fNIRS features showed limited discriminative power. Early fusion of EEG with ECG, EMG, and fNIRS improved classification performance, giving perfect accuracies with all the feature sets. These highlights the superiority of multimodal fusion over unimodal analysis, underscoring the potential of hybrid physiological signal integration for reliable emotion recognition systems.