MÉTODO DE EXTRACCIÓN Y SELECCIÓN DE CARACTERÍSTICAS PARA LA IMPLEMENTACION DE UNA INTERFAZ CEREBRO-COMPUTADORA EN DETECCIÓN DE EMOCIONES

The absence of a reliable method for emotion detection by using electroencephalographic (EEG) signals obtained by portable Brain-Computer interfaces (BCI) stills without a clear answer from the multiple research in this topic. The main goal is propose an approach to obtain parameters (attributes) for the classification of patterns related with the activity of EEG during an affective stimulus, responding to the current need to build computational systems that improve the understanding of emotional human behavior and its application in other fields of research. The approach consists on discrete dyadic Wavelet decomposition as the extraction of representative data applied over the pre-proccessed EEG from one sensor, followed by the features selection step, and then the supervised learning of patterns with different types of algorithms, in order to compare the results and set the conditions with the best performance. A public access database was used to test the proposed method. The results of the classification threw a mean accuracy of 66.5% for the arousal axis and 68.1% for the valence axis, with the Naïve-Bayes classifier and the individual gain information of the features. In summary, the approach is potentially useful with a non-exhaustive training step by using Naive Bayes classifier, which is an advantage during the online emotion detection tests, and remarks the viability of use a minimal number of sensors, replacing them by feature combinations. Keywords: Affective computing, Interface, Brain device, Human-Computer Interaction, Detection, Portable.