Emotion profiling using deep learning

• Main Author: Arora, Srishti
• Other Authors: Guan Cuntai
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/166126
• Institution: Nanyang Technological University

Individuals diagnosed with distress disorders exhibit heightened levels of negative affective factors that stem from underlying motivational systems associated with threat, safety, and loss of rewards. Emotion Regulation Therapy (ERT) is a psychotherapeutic approach that seeks to enhance the patient’s awareness of their motivational processes, develop their emotional regulation capacities, and facilitate new contextual learning. Neurofeedback training usingEEGhas showna potential to be employed as a therapeutic tool for emotion regulation and thereby improve the efficiency of Emotion Regulation Therapy. This technique involves providing real-time feedback to individuals about their brainwave activity while they engage in specific emotion regulation tasks. By using this feedback, individuals can learn to regulate their brainwave activity and, in turn, their emotional responses. Neurofeedback training relies on Emotion Recognition Systems for identifying the brainwave patterns associated with different emotional states and studying the neural mechanisms involved in emotion regulation. Motivated by the utility of emotion recognition in the treatment of mental health issues, this project aims to explore how emotion regulation therapy or music therapy can be made more effective with the help of EEG. The project involves the development of an emotion recognition system for studying how brain activity, recorded using EEG, varies with emotions. This is achieved by collecting data through experiments that attempted to induce different states of emotions using various music stimuli. This data is then analyzed to ensure emotions are induced and brain activities corresponding to different stimuli are distinguishable. Emotion states are then classified by a CNN-based deep neural network and a maximum accuracy of 69% is achieved.