Electroencephalography-Based Emotion Recognition Using Auditory Stimulation for Affective Brain-Computer Interfaces
Charoenporn Bouyam1, Nannaphat Siribunyaphat1,2, Si Thu Aung3
1School of Informatics, Walailak University, Nakhon Si Thammarat 80160, Thailand.
Sensors (Basel, Switzerland)
|August 13, 2026
Summary
Electroencephalography (EEG) can recognize emotions using auditory stimuli, but faces challenges with individual differences. Few-shot learning significantly improved accuracy in subject-independent emotion recognition systems.
Area of Science:
- Neuroscience
- Affective Computing
- Machine Learning
Background:
- Electroencephalography (EEG)-based emotion recognition shows promise for affective brain-computer interfaces (BCIs).
- Significant inter-subject variability in EEG signals hinders the generalizability of current emotion recognition models.
- Developing robust and personalized BCI systems requires addressing these individual differences.
Purpose of the Study:
- To propose and evaluate an EEG-based framework for emotion recognition within a valence-arousal model using auditory stimulation.
- To systematically assess the performance of different EEG features (DWT, FC, EC) and machine learning classifiers.
- To investigate the effectiveness of subject-dependent, subject-independent, and few-shot subject-adaptation protocols.
Main Methods:
- EEG data were collected during auditory stimulation (instrumental melodies) designed to elicit specific emotional states.
- Features extracted included Discrete Wavelet Transform (DWT), Functional Connectivity (FC), and Effective Connectivity (EC).
- Five machine learning classifiers were used, and performance was evaluated using subject-dependent, leave-one-subject-out (LOSO), and few-shot adaptation protocols.
Main Results:
- Subject-dependent classification achieved high accuracy, with DWT yielding the best performance (0.88).
- Subject-independent (LOSO) evaluation showed near-chance performance (0.23-0.30), confirming substantial inter-subject variability.
- Few-shot subject adaptation significantly improved subject-independent performance, reaching 0.77 accuracy with FC using 75% calibration data.
Conclusions:
- EEG-based emotion recognition is feasible using auditory stimulation within defined affective states.
- Addressing inter-subject variability through few-shot learning is crucial for developing personalized affective BCIs.
- The proposed framework and findings lay the groundwork for future advancements in adaptive BCI systems.
