A research team at Korea Advanced Institute of Science and Technology (KAIST) said it has developed Hi-COVIDNet, a big data and artificial intelligence platform that can predict the number of Covid-19 inflows from abroad.

KAIST Professor Lee Jae-gil (seated at second from right) and his research team have developed a model that predicts the number of the Covid-19 virus inflows from abroad. (KAIST)

The technology developed by the research team, led by Professor Lee Jae-gil, applies AI to big data, such as the number of confirmed and fatalities, the frequency of searching for Covid-19-related keywords, the number of roaming customers entering Korea from overseas countries, and the number of daily flights to Korea from foreign countries, to predict the amount of virus inflows for the next two weeks.

Researchers used the number of confirmed cases and deaths reported as a baseline when calculating the risk of Covid-19 in overseas countries. Since these numbers depend on the number of diagnostic tests, however, they also used Covid-19-related keyword search frequency as input data to calculate the country risks in real-time.

Since the real-time number of entrants is confidential and not disclosed to the public, the team also derived such a number from the daily number of flights arriving in Korea and the number of roaming customers entering Korea.

“We had to consider geographic relevance between countries as an important factor in predicting the number of confirmed imported cases,” the team said. “This is because the outbreak of Covid-19 in a particular country can more easily spread to neighboring countries, and distance among countries also affected their exchanges.”

To solve this problem, the research team accurately predicted the number of confirmed cases of foreign inflows from each continent. After developing the platform, the team predicted the number of confirmed imported cases over the next two weeks through Hi-COVIDNet. As a result, it confirmed that this model has up to 35 percent higher accuracy than the existing time-series data-based predictive machine learning or deep learning-based models.

“As the number of confirmed cases of Covid-19 increases rapidly, the risk of overseas inflows spreading to the local communities always remains,” it said. “We expect the accurate prediction technology will be very useful for estimating the expansion of quarantine facilities and controlling immigration from high-risk nations.”

The KAIST team plans to present its research during the ACM KDD 2020 conference on Aug. 24.

Copyright © KBR Unauthorized reproduction, redistribution prohibited