Multi-Classifier Fusion Based on MI-SFFS for Cross-Subject Emotion Recognition.

Area of Science:

• Affective computing
• Neuroscience
• Machine learning

Background:

• Emotion recognition using Electroencephalography (EEG) is crucial for understanding human affective states.
• Cross-subject emotion recognition presents challenges due to inter-individual variability in EEG signals.
• Improving feature generalization is key to enhancing the accuracy of EEG-based emotion recognition systems.

Purpose of the Study:

• To propose and evaluate a novel Multi-Classifier Fusion method for improving cross-subject emotion recognition accuracy.
• To enhance the generalization of extracted EEG features for more robust emotion classification.
• To investigate the effectiveness of integrating mutual information with sequential forward floating selection (MI_SFFS) for feature selection.

Main Methods:

• Utilized the DEAP dataset, extracting features from 15 EEG channels within a 10-second time window.
• Implemented a novel feature selection technique combining mutual information (MI) and sequential forward floating selection (SFFS).
• Employed a Multi-Classifier Fusion approach, using Support Vector Machine (SVM), k-nearest neighbor (KNN), and Random Forest (RF) classifiers, with their output probabilities serving as weighted features.

Main Results:

• Achieved cross-subject classification accuracies of 0.7089 (SVM), 0.7106 (KNN), and 0.7361 (RF).
• Demonstrated the feasibility of the proposed model by effectively combining classifier outputs as weighted features.
• Validated the model's performance using leave-one-out cross-validation.

Conclusions:

• The proposed Multi-Classifier Fusion method, incorporating MI_SFFS feature selection, significantly enhances cross-subject emotion recognition accuracy from EEG signals.
• The approach of splicing classifier output probabilities as weighted features proves effective for improving model generalization.
• This research contributes a viable method for more accurate and reliable EEG-based emotion recognition in affective computing.