Researchers Develop AI Model That Detects Emotions From Brain Signals
Researchers at Innopolis University, Saint Petersburg Electrotechnical University "LETI," and Jadavpur University in India have developed a hybrid neural network that identifies stress, joy, and other human emotions from electroencephalogram (EEG) data. The model achieves accuracy of up to 99.99%.
The researchers tested the new architecture on three EEG datasets. The system achieved its highest accuracy when identifying calmness, stress, and joy, while recording its lowest performance in distinguishing negative, neutral, and positive emotional states, where accuracy reached 96.49%.
The architecture combines two neural network branches: a temporal branch that analyzes EEG signals and a spectral branch that extracts features using a convolutional neural network.
The findings were published in the journal Scientific Reports.
