Kansas State University Undergraduate Research Conference

Wearing the Inside Out: Using Long Short-Term Memory Networks and Wearable Data to Identify Human Emotions

Studying emotions may sound unusual in computer science, a field based on quantifiable data and rationality. Contrary to belief, studies have shown any decision is highly dependent on emotional input. To improve human-computer interaction, it is crucial to improve our understanding of human emotions and teach machines to identify them. With large amounts of information streaming available from our environment, identifying our current emotional state becomes challenging, even at the individual self-level. This project aims to identify indicative emotional temporal data from wearable devices. Using brain activity data from an EEG and smart watches that record data, such as heart-beat, physical motions and glucose-levels, we hope to find a correlation that will enable us to train a neural Long Short-Term Memory Network (LSTMN) that classifies the temporal physical-state data into the emotional state of the subject. LSTMNs allow the use of previous long- and short-term data points, expanding our understanding of what our body is telling us about our psyche.