An Ensemble Deep Learning Approach for EEG-Based Emotion Recognition Using Multi-Class CSP.

Area of Science:

• Neuroscience
• Computer Science
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) are advancing, with emotion recognition from EEG signals being a key area.
• Previous research often overlooked critical spatial-temporal EEG characteristics, limiting accuracy.
• Accurate emotion detection is vital for developing more intuitive and responsive BCIs.

Purpose of the Study:

• To present a novel approach for classifying emotions (positive, negative, neutral) using EEG signals.
• To address the limitations of prior studies by incorporating spatial-temporal EEG features.
• To develop a high-accuracy emotion recognition system for BCI applications.

Main Methods:

• Collected a custom dataset of EEG signals from 16 participants experiencing induced emotional states via music.
• Employed Multi-class Common Spatial Pattern (MCCSP) for EEG signal processing.
• Utilized an ensemble model with three Convolutional Neural Network (CNN) autoencoders for classification.

Main Results:

• Achieved a classification accuracy of 99.44 ± 0.39% for positive, negative, and neutral emotional states.
• The proposed method demonstrated superior performance compared to previous studies.
• Effectively utilized spatial-temporal EEG characteristics for enhanced emotion recognition.

Conclusions:

• The developed method shows significant promise for real-world BCI applications.
• High accuracy in emotion detection provides a reliable foundation for future BCI development.
• The approach validates the importance of spatial-temporal EEG features in emotion recognition.