New computer model tracks cell-phone data to predict COVID spread

Last year, Arti Ramesh and Anand Seetharam — both assistant professors in the Department of Computer Science at Binghamton University’s Thomas J. Watson College of Engineering and Applied Science — published several studies that used data-mining and machine learning models to respond to the COVID-19 pandemic.

One study used coronavirus data collected by Johns Hopkins University to show how and where infections could spread on a global scale. Another study used anonymous cell-phone data to track how residents of Rio de Janeiro traveled throughout the city before, during and after its strictest lockdown protocols.

In their latest research, Ramesh and Seetharam have blended the two ideas into a new algorithm that narrows the geographic scope of their COVID predictions, making it more useful for regional and local officials looking to curb the spread of the virus.

Returning their attention to Brazil’s second-largest city, they correlated cell-phone data with coronavirus infection rates in Rio’s municipal districts to build a mathematical model that predicts how cases would change in the next seven days for the different municipalities. The projections are based on current COVID rates combined with mobility trends to and from the districts with the most infections.

“This is one of the first studies that has quantified mobility in a manner that it can be used to demonstrate how cases are going to spread,” Seetharam said. “It’s not just the number of cases in a particular region that contributes to future cases in that region.”

“These municipalities are units where the data can be recorded and collected easily,” Ramesh said. “Every city or jurisdiction has these kinds of units where the numbers of cases are recorded separately or that are governed separately from other municipalities. That action is very helpful in converting the observations we are making into policies.”

Ramesh and Seetharam believe their prediction model could be mapped onto larger regions such as counties, states or even countries, giving political leaders and health officials the ability to better predict where COVID could spread the fastest.

Short-term solutions might include partial lockdowns of certain neighborhoods or towns, such as New York restricting movement into and out of New Rochelle last March when the virus arrived in the U.S.

“We wanted to build a model that can be translated easily into policy, one with subregions that are managed through independent jurisdictions,” Ramesh said. “That way, someone can actually impose a lockdown on that level, as well as collect data about cases and predict on that level.”

Longer-term policies also could be made using the Binghamton team’s prediction algorithm.

“If you know there are a lot of cases in a particular region, you could think of a future world where this model could be used for some kind of city planning,” Seetharam said. “You could create traffic flows in such a way that people try to avoid that region because of the inconvenience involved. That would lessen the spread of infection. It need not be anything invasive, but minor tweaks could be beneficial.”

Ramesh and Seetharam’s study is called “Mobility-aware COVID-19 Case Prediction using Cellular Network Logs.”