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

Abstract

In recent years, emotion recognition using brain-computer interface (BCI) systems has gained substantial attention. Existing models are typically implemented in either offline (batch) or online (streaming) modes. While batch processing approaches generally achieve higher classification accuracy, they are limited by slow processing speed. In contrast, stream processing approaches offer real-time performance but often compromise accuracy. To address this trade-off, we propose a hybrid batch-streaming framework that integrates the strengths of both paradigms while alleviating their individual limitations. The architecture, features a probabilistic intelligent switching mechanism that estimates the reliability of the streaming module based on its historical performance. This reliability measure dynamically determines the probability of selecting outputs from either the batch or streaming unit. The proposed framework is evaluated on three benchmark datasets (DEAP, AMIGOS, and SEED) achieving classification accuracies of 85%, 94%, and 74%, respectively. Also, experiments were conducted to investigate the performance of the switch mechanism and performance of system components against concept drift. Experimental results demonstrate that our method effectively balances classification accuracy and computational efficiency. It is expected that in the future, such hybrid ideas are widely used in feedback - based systems.