CNN-XGBoost fusion-based affective state recognition using EEG spectrogram image analysis.

Area of Science:

• Neuroscience and Artificial Intelligence
• Brain-Computer Interfaces
• Affective Computing

Background:

• Human emotion recognition from brain signals is challenging.
• Existing methods often rely on traditional feature extraction and fusion techniques.
• There is a need for advanced methods to accurately classify emotional dimensions.

Purpose of the Study:

• To propose a novel signal spectrogram image-based CNN-XGBoost fusion method for emotion recognition.
• To accurately classify three dimensions of human emotion: arousal, valence, and dominance.
• To compare the proposed method against state-of-the-art feature fusion approaches.

Main Methods:

• EEG signals from the DREAMER dataset were converted into spectrogram images using Short-Time Fourier Transform (STFT).
• A 2D Convolutional Neural Network (CNN) was employed to extract features from spectrogram images.
• Extreme Gradient Boosting (XGBoost) classifier was applied to the extracted CNN features for emotion classification.

Main Results:

• The proposed CNN-XGBoost fusion method achieved high accuracy: 99.712% for arousal, 99.770% for valence, and 99.770% for dominance.
• The method demonstrated superior performance compared to feature fusion-based SVM and XGBoost approaches.
• Spectrogram image-based feature extraction proved more effective than traditional signal processing techniques.

Conclusions:

• The CNN-XGBoost fusion method using EEG spectrograms is highly effective for human emotion recognition.
• This approach offers a significant advancement over existing feature-level fusion techniques.
• The findings highlight the potential of deep learning on image representations of brain signals for affective computing.