EEG-based emotion recognition with manifold regularized extreme learning machine.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Electroencephalography (EEG) signals offer a non-invasive method for studying human emotional states.
• Previous methods for EEG-based emotion recognition have limitations in capturing complex data structures.

Purpose of the Study:

• To propose a novel algorithm, manifold regularized extreme learning machine (MRELM), for enhanced EEG-based emotion recognition.
• To evaluate the performance of MRELM in classifying emotional states (positive, neutral, negative) evoked by visual stimuli.

Main Methods:

• Utilized differential entropy features across five frequency bands (delta, theta, alpha, beta, gamma) from EEG data.
• Developed and applied the manifold regularized extreme learning machine (MRELM) algorithm.
• Compared MRELM performance against Generalized Extreme Learning Machine (GELM) and Support Vector Machine (SVM).

Main Results:

• MRELM achieved an average accuracy of 81.01%, outperforming GELM (80.25%) and SVM (76.62%).
• High-frequency bands (beta and gamma) demonstrated superior relevance for detecting emotional state transitions compared to lower frequency bands.
• MRELM showed robust performance even when training and testing datasets were from different sessions.

Conclusions:

• MRELM is an effective and competitive model for recognizing human emotional states from EEG data.
• The findings underscore the importance of high-frequency EEG bands in understanding emotional dynamics.
• The proposed algorithm offers a promising approach for real-world applications in affective computing.