Real-time emotion recognition based on EEG signals using a hybrid batch-stream architecture

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Emotion recognition using Brain-Computer Interface (BCI) systems is gaining traction.
• Current BCI models face a trade-off between offline (batch) processing accuracy and online (streaming) processing speed.
• Batch processing offers high accuracy but is slow; streaming offers real-time performance but lower accuracy.

Purpose of the Study:

• To develop a hybrid framework integrating batch and streaming BCI paradigms.
• To address the accuracy-speed trade-off in real-time emotion recognition.
• To introduce a probabilistic intelligent switching mechanism for dynamic output selection.

Main Methods:

• Proposed a hybrid batch-streaming framework for BCI emotion recognition.
• Implemented a probabilistic intelligent switching mechanism based on historical streaming module performance.
• Evaluated the framework on DEAP, AMIGOS, and SEED benchmark datasets.

Main Results:

• Achieved classification accuracies of 85% (DEAP), 94% (AMIGOS), and 74% (SEED).
• Demonstrated effective balancing of classification accuracy and computational efficiency.
• Investigated the performance of the switching mechanism and system components against concept drift.

Conclusions:

• The proposed hybrid framework successfully mitigates the limitations of individual batch and streaming BCI approaches.
• The intelligent switching mechanism dynamically optimizes performance based on reliability.
• This hybrid approach is expected to be influential in future feedback-based BCI systems.