AI-Powered Emotion Detection from EEG Signals

Abstract

Emotion recognition from brain activity has emerged as an important research area in affective computing, with electroencephalography (EEG) providing a direct physiological signal for identifying emotional states. This study presents a controlled experimental comparison of three deep learning architectures—one-dimensional Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU)—for classifying EEG-derived features into negative, neutral, and positive emotional states. A publicly available EEG Brainwave dataset containing 2,132 samples recorded using a 14-channel Emotiv EPOC headset was used for experimentation. The same preprocessing procedure, data split, training configuration, and evaluation criteria were applied to all three architectures to provide a consistent basis for comparison. The models were evaluated using accuracy, precision, recall, F1-score, and confusion-matrix analysis. Experimental results show that the GRU achieved the highest test accuracy of 98.1%, followed by LSTM with 96.2% and CNN with 95.1%. The GRU also achieved the highest macro-averaged precision, recall, and F1-score, demonstrating its effectiveness in capturing sequential patterns within EEG-derived features. The findings indicate that gated recurrent architectures can provide strong classification performance while maintaining relatively lower architectural complexity than graph- and Transformer-based approaches. The study highlights the potential of GRU-based models for resource-efficient EEG emotion-recognition systems and provides a controlled benchmark for comparing commonly used deep learning architectures.