Enhanced Single-Subject Eeg Emotion Recognition Using Deep Neural Architectures And Representation Learning

Electroencephalography (EEG)-based emotion recognition has become a vital element of both emotional computing and custom human-computer interaction systems. The present methods for recognising emotional states through EEG signals depend on either manual feature extraction methods or basic machine learning systems, which restrict their performance and make it impossible to use them in real-time personalised settings. The research presents a single-subject EEG emotion recognition system that uses deep learning for developing a system that connects spatial-spectral feature extraction with temporal affective sequence modelling. The system uses a combined Convolutional Neural Network and Long Short-Term Memory architecture to learn emotional state neural patterns, which can differentiate between multiple emotional states. The study assessed performance using a within-subject research design, which required both training and testing data to come from the same subject to take advantage of stable brain activity patterns. The research team analysed multiple deep learning architectures, which included CNN LSTM, BiLSTM, and CNN-LSTM, through standard performance assessment methods. The hybrid model achieved its best result by attaining 98.32% overall accuracy and 0.983 Macro-F1 score, which exceeded the performance of standalone CNN, BiLSTM and LSTM models. The study results show that hybrid deep neural architectures work well for personalised EEG-based emotion recognition, while they also have potential to be used in specific user emotional computing applications, which include neurofeedback and adaptive gaming, cognitive monitoring and human-computer interaction. The findings provide a strong baseline for future work involving real-time implementation and subject-generalising strategies.