Clasificación de Emociones a partir de Señales de EEG utilizando Supervectores de Características y Selección Óptima de Canales

Affective Computing aims to endow systems with the ability to recognize and respond to human emotional states. This work addresses emotion recognition from electroencephalography signals by evaluating feature extraction strategies in the time, frequency, and time–frequency domains. The extracted features are integrated into supervectors, followed by dimensionality reduction using principal component analysis, and subsequently classified using support vector machines with polynomial and radial basis function kernels. Additionally, an exhaustive electrode selection procedure is implemented to identify optimal subsets of channels. Experiments were conducted on the SEED dataset (62 channels). The results show consistent performance improvements when combining supervectors, principal component analysis, and support vector machines, as well as when using a reduced number of relevant channels, achieving a maximum accuracy of 0.9067. This approach provides a solid, interpretable, and low computational-demand baseline for affective brain–computer interface systems, in which optimizing model accuracy without compromising the computational efficiency required for real-time applications is essential.