Emotion recognition from multimodal biosignals: supervised and unsupervised machine learning approaches based on EEG...

Area of Science:

• Psychology
• Computer Science
• Neuroscience

Background:

• Artificial intelligence (AI) and machine learning (ML) are increasingly used in psychology to analyze complex behavioral and physiological data in educational and clinical settings.
• Limited evidence exists on using multimodal physiological signals to characterize cognitive and emotional processes in real-world educational environments.
• This study investigates the potential of ML for analyzing physiological responses to emotional stimuli in higher education.

Purpose of the Study:

• To explore supervised and unsupervised ML approaches for analyzing multimodal physiological responses.
• To assess the capacity of physiological signals to characterize cognitive and emotional processes in an educational context.
• To evaluate the predictive contribution of physiological, subjective, and sociodemographic variables.

Main Methods:

• Recruited 48 university students and lecturers.
• Recorded electroencephalography (EEG) and galvanic skin response (GSR) while participants observed emotional avatars.
• Extracted event-related potential (ERP) features and applied Random Forest, k-means clustering, and principal component analysis (PCA).

Main Results:

• Sociodemographic variables showed the highest predictive performance, followed by subjective workload.
• Physiological features had limited discriminative power.
• Cluster analysis revealed overlapping response profiles, and PCA showed limited group separation, suggesting continuous patterns of variability.

Conclusions:

• ML offers opportunities and limitations for analyzing physiological data in educational settings.
• Physiological measures provided complementary information but had limited predictive capacity alone.
• Future research requires larger samples and advanced multimodal modeling for improved interpretability and predictive value of physiological signals.