UC-20 Analyzing Concentration Levels in Online Education Using Machine Learning

These past few years have introduced the most important time in history to study new faucets of online learning. Due to COVID\u27s impact, online learning became a staple in millions of homes across the country. Guided by the research question, “Can a machine learning model be created and trained to detect student concentration level based on eye and facial data?”, we set out to contribute to society’s understanding of online learning under the guidance of Dr. Ying Xie and Dr. Linh Le. Our process involved recording ourselves participating in online classes to garnish eye and facial data. Each student recorded data where they were staring directly at the screen and focused while also recording data from when the student was distracted and looking elsewhere. Next, deepfake technology was used to swap our faces with celebrity images to protect our privacy while also generating a large pool of data. Finally, we created, trained, and tested different machine learning models to try and find settings best suited for our needs. Preliminary results indicated our top constructed models hovering around 65-75% accuracy. This was all completed using open source software, and we believe that if access was granted to even more accurate eye tracking/deepfake technologies, the accuracy of the model would increase even further. Overall, our results indicate that it would be possible to develop a model that could indicate whether a student was concentrated or distracted based on eye and facial data while implementing a process that could protect the student’s identity, which could be useful data for either the student to review their own performance in class or for a teacher to see at what point students become distracted during an online session.Advisors(s): Dr. Ying Xie (Co-sponsor) Dr. Linh Le (Co-sponsor)Topic(s): Data/Data AnalyticsIT 498