Introduction: Emotion identification is an important part of human intelligence and it plays an important role in interpersonal communication. Unlike the human communication, identifying the emotions of a person is challenging when it comes to human machine communication. In recent years, studies based on human brain emerged computer technologies to detect emotions, especially noninvasive EEG techniques had been popular in the research area due to its affordability, simplicity and easiness to use. It gives more information which cannot be extracted from verbal, facial or any other physiological aspects and, currently this technology is deployed in various domains.

Human emotions are complex and they are produced due to various chemical activities inside the human brain. The most challenging part is that the emotions are relative to person and it cannot be defined quantitatively. Therefore, as a common and fair measure most of the studies on emotion detection have been done by considering the arousal and valance of the motions.

Methods and Results: The aim of this study is to construct a machine learning model to identify emotions based on their valance. For that, the EEG recordings from 10 subjects, equally in both genders, were recorded. BioRadio device with T3, T4, F3 and F4 electrodes were used to create the dataset and audio-visual stimuli was used for the elicitation. These four electrodes were placed on the subject’s scalp and a three minutes calming video was played initially to bring the subject’s emotion to neutral. Then a positive or negative video was played for five minutes. During this session EEG data was recorded and at the end the subject is also asked to rate his emotions where he reaches neutral and positive or negative. The data collection was done in two separate days for a single subject and each day, 3 and 4 videos randomly from each class was played respectively. Different features from time and frequency domains are being investigated and it is more focused on signal processing techniques to identify the most discriminative features. All the analysis were done on 1 minute length samples which were obtained by trimming the collected samples. Support Vector Machines with different kernels will be used to classify emotions into negative, positive and neutral classes, based on their valance.

Keywords: Emotion Detection, Signal Processing, Electroencephalogram