BiLSTM-Based Human Emotion Classification Using EEG Signal

Emotion recognition using electroencephalography (EEG) signals has garnered significant attention due to its applications in affective computing, human-computer interaction, and healthcare. This study employs a Bidirectional Long Short-Term Memory (BiLSTM) network to classify emotions using EEG data from four well-established datasets: SEED, SEED-IV, SEED-V, and DEAP. By leveraging the temporal dependencies inherent in EEG signals, the BiLSTM model demonstrates robust learning of emotional states. The model achieved notable classification accuracies, with 92.30% for SEED, 99.98% for SEED-IV, 99.97% for SEED-V, and 88.33% for DEAP, showcasing its effectiveness across datasets with varying class distributions. The superior performance on SEED-IV and SEED-V underscores the BiLSTM's capability to capture bidirectional temporal information, which is crucial for emotion recognition tasks. Moreover, this work highlights the importance of utilizing diverse datasets to validate the generalizability of EEG-based emotion recognition models. The integration of both dimensional and discrete emotion models in the study demonstrates the framework's flexibility in addressing various emotion representation paradigms. Future directions include optimizing the framework for real-world applications, such as wearable EEG devices, and exploring transfer learning techniques to enhance cross-subject and cross-cultural adaptability. Overall, this study advances EEG-based emotion recognition methodologies, establishing a robust foundation for integrating affective computing into various domains and paving the way for real-time, reliable emotion recognition systems.