Emotional eye movement analysis using electrooculography signal

Abstract

In this study, for recognition of (positive, neutral and negative) emotions using EOG signals, subjects were stimulated with audio-visual stimulus to elicit emotions. Hjorth parameters and Discrete Wavelet Transform (DWT) (Haar mother wavelet) were employed as feature extractor. Support Vector Machine (SVM) and Naive Bayes (NB) were used for classifying the emotions. The results of multiclass classifications in terms of classification accuracy show best performance with the combination DWT+SVM and Hjorth+NB for each of the emotions. The average SVM classifier's accuracy with DWT for horizontal and vertical eye movement are 81%, 76.33%, 78.61% and are 79.85%, 75.63% and 77.67% respectively. The experimental results show the average recognition rate of 78.43%, 74.61%, and 76.34% for horizontal and 77.11%, 74.03%, and 75.84% for vertical eye movement when Naive Bayes group with Hjorth parameter. Above result indicates that it has the potential to be used as real-time EOG-based emotion assessment system.