Comparación de métodos de extracción de características en el dominio del tiempo, la frecuencia y tiempo-frecuencia usando datasets públicos para reconocimiento de emociones usando OpenBCI.

Abstract

In the literature, several works have focused on emotion recognition using psychological cues such as skin temperature, heart rate, and brain signals. The study of brain signals has gained much popularity in recent years, as it is unlikely to fake brain activity read through EEG signals. Analysis of EEG signals involves two main processes: feature extraction and classification. This study's main objective is to compare feature extraction methods in the time, frequency, and time-frequency domains using the SEED and DEAP public datasets for emotion classification in EEG signals. We analyzed different characteristics of the three domains; the efficiency of the classifiers' input vectors was validated using the SVM, KNN, and ANN classification algorithms that were selected for being more used and presenting better performances in the state_of_the_art. The proposed model can identify emotions in four quadrants of the valence-arousal representation space using the DEAP dataset and emotions according to their valence, such as neutral, sad, fear, and happiness, in the SEED dataset. We achieved a classification accuracy of 66.20% using time-domain features and the k-NN classifier and 66.0% using frequency domain features and the SVM classifier for the DEAP dataset, with the frequency domain the one with the lower computational cost. We obtained an accuracy of 67.40% for the SEED dataset using time domain features and the SVM classifier.