Reconocimiento de estrés emocional en el plano Valencia-Arousal usando señales EEG y aprendizaje automático.

Abstract

In this study, we recognized which characteristics contribute to identifying stress using electroencephalographic (EEG) signals collected by a BCI and labeled in the valence-arousal plane. It will begin with the use of the DEAP database [1]. defined for their recognition in the valence-arousal plane. This step is critical since what is considered stress is labeled from this information according to the range of values of these indicators. The characteristics extraction is done through different techniques proposed in [3] and applied to the DEAP data set to obtain attributes in the domain of time, frequency, and time-frequency. With these characteristics together with the raw data, different input vectors are formed to enter them into various configurations of Convolutional Networks (CNN) with different compositions in each of their layers: input, hidden and output. Combinations that produce the best performance in terms of their accuracy (accuracy) will be obtained experimentally. Finally, this work also generates its own data. For this task, stimulation methods are used to cause stress in the participants during the experiments. For this, the method of applying arithmetic problems with a limited time to solve them is evaluated [5] [6]. With DEAP and the own database of stress stimuli, 84% and 69% were obtained respectively in the best combinations of the methods used.