Scientists from Russia and India have developed a hybrid neural network capable of identifying stress, joy, and other emotional states by analyzing electroencephalography (EEG) signals with an accuracy of up to 99.99%.
This was reported by the TASS news agency, citing the press service of the Russian Innopolis University.
The university said: "Researchers from Innopolis University, Petersburg State University of Electrical Engineering, and Jadavpur University in India have created a hybrid neural architecture that analyzes electroencephalograms to identify stress, joy, and other emotional states. This technology could be useful for developing mental health monitoring systems and adaptive human-computer interfaces."
The new model was tested on three standard EEG datasets and was able to identify states of calm, tension, and joy with 99.99% accuracy. Researchers plan to adapt the model to recognize emotions in real time using streaming EEG data.
A spokesperson for Innopolis University explained that the new model outperformed or matched previously published results when tested on the same datasets. For example, other models achieved a best accuracy of 98.55% in a dataset encompassing calm, stress, and joy, while the new model achieved 99.99% accuracy. In the dataset for negative, neutral, and positive emotions, accuracy increased from 95.99% in the best previous models to 96.49% using the new model.
Dmitry Kaplun, chief mathematics programmer at the Artificial Intelligence Research Center at Innopolis University, said that monitoring mental health through emotion recognition plays an important role in developing personalized healthcare systems and in the early detection of depression, anxiety, and stress-related disorders.
He added: “If current methods for identifying emotions through brain mapping using artificial intelligence require manual selection and adjustment of features, suffer from limitations in generalization to different datasets, as well as high computational costs, our new model overcomes these limitations, as it enables faster and more accurate results, while reducing computational costs.”
The results of the experiments were published in the scientific journal Scientific Reports.
