Emotion recognition using brian signals based on time-frequency analysis and supervised learning algorithm

Abstract

Over the years many groundbreaking research involving Brain Computer Interface (BCI), has been conducted in order to study emotions of human beings, to build better-quality human-machine interaction systems. On the other hand, it is also quite possible to log the activities of brain in real-time and then use it to distinguish patterns related to emotional status. BCI creates a mutual understanding between the users and its environment for measuring emotions through brain activities. Electroencephalogram (EEG) is a well-accepted method to measure the brain activities. Once the system records the EEG signals, we analyze and process these activities to distinguish different emotions. Previous researchers used standard and pre-defined methods of signal processing area with fewer channels and participations to record their EEG signals. In this thesis, a novel method was proposed that extracted features from EEG signals based on time-frequencies analysis and supervised learning algorithm was used to classify different emotional states. Our proposed method provides 92.36% accuracy by using a benchmark dataset, where 32 participants were used to carry out this experiment.