DENTON (UNT), Texas — A class project turned into much more as 30 teams of University of North Texas students compete internationally to create the most accurate model of the spread of COVID-19.
The students, all currently taking artificial intelligence classes with assistant professor Mark Albert and research assistant professor Ting Xiao in the College of Engineering, joined more than 400 teams from around the world using real-world data in an attempt to predict the number of infections and fatalities for each region of the world as part of the Kaggle COVID-19 Global Forecasting challenge. Before the April 15th deadline, teams were given the option of submitting their model predictions for April 16 to May 14, which many of the UNT teams did.
Kaggle, an online resource for computer enthusiasts with more than 19,000 public datasets and frequent challenges, is hosting a variety of global competitions aimed at better understanding COVID-19 through data analysis and natural language processing. Other challenges listed on their site include predicting unit sales for large department stores, predicting housing prices and predicting which tweets in a given set are about real disasters and which are not.
“Dr. Xiao and I each regularly schedule Kaggle prediction competitions later in the semester for our AI-related classes. Sometimes we set it up so the students only compete with each other on a data set of our choosing, but this time we decided to have them join a global competition given the circumstances,” Albert said. “Some students submitted a very basic model, while others researched the discussion boards and used state-of-the-art recurrent neural network models tuned for this task.”
Initially, the competition had 472 teams, with six UNT teams in the top 50. As of April 28th, five students as part of UNT’s team “CSCE-5300-kaggle-challenge” were in 26th place out of 313 teams still participating in the challenge – rare for a team of students in their first Kaggle competition.
While the models cannot be updated once they are submitted, the models and rankings are re-evaluated every day until May 14 as global conditions change.
Xiao said she loves the challenges because they often “light a fire” under students who would normally excel, pushing them beyond the classroom. Over time, she said, that drive pushes them to the top of their chosen fields.
Students Abolfazl Meyarian and Havish Nallapareddy, both on the current top UNT team, said they were excited to be part of the challenge not only to apply what they learned in class to a real, global issue, but also because they both feel science can bring solutions that provide hope in a crisis. After building their model – a task they completed in a week to meet the competition deadline – they realized they had predicted the path of the virus with a fair amount of accuracy. This made them even more optimistic about how artificial intelligence could help plan for future events.
“Competitions like this provide a clear sense of achievement that tests and quizzes don't,” Albert said. “Instead of an instructor grading right versus wrong, you have a score telling you how good your predictions are, and the better your model the better the score.”
For Meyarian and Nallapareddy, the competition not only pushed them, it helped them decide what they want to do in the future. Realizing the effect of extracting data to help improve people’s lives, they both decided they want to make their marks in both artificial intelligence and health.