Abstract
The field of emotion recognition aims to automatically quantify human emotional states based on behavioural and physiological data. Interesting applications of this area of study include the early detection and relieve of strong negative emotions, and the diagnosis of neurological disorders that affect people’s emotional well being. Much of the current emotion recognition research aims to differentiate from 3 emotional states, positive, negative, or neutral, and further work is needed to classify more nuanced emotional states. In addition, there is growing interest for classifying emotions based on wearable devices that are available beyond laboratory settings. My research develops a convolutional neural network algorithm that is able to estimates participants’ ratings of 9 distinct emotions based on EEG, EDA, and BVP data collected with commercial-grade devices. My algorithm performs a regression estimate that is within 0.30 of participants’ self-reported emotions on a 0 to 4 scale.