“Our model is the first that uses coronavirus data and integrates two areas: machine learning and standard epidemiology,” explains Raj Dandekar, PhD candidate studying civil and environmental engineering. With mechanical engineering professor George Barbestathis, Dandekar has spent the last few months developing it as part of the final project in class 2.168 (learning machines).
Most of the models used to predict the spread of a disease follow what is known as the SEIR model, which makes people “susceptible,” “exposed”, “infected” and “recovered”. . Dandekar and Barbastathis extended the SEIR model by training the neural network to capture the number of infected individuals, and therefore no longer spreading the infection to others.
The model finds that in places like South Korea, where there was immediate government intervention in implementing strong quarantine measures, the virus spread more quickly. “Effective reproduction numbers” of more than one residue, which means the virus continues to spread rapidly, in slow places to implement government interventions such as Italy and the US.
Machine learning algorithms suggest that plateaus for both Italy and the United States will reach elsewhere on April 15–20, with the quarantine measures currently in place. This prediction is similar to other estimates such as the Institute for Health Metrics and Evaluation.
“Our model suggests that quarantine bans are successful in achieving effective reproduction numbers from one to the smallest,” says Barberathis. “This corresponds to the point where we can level the curve and see fewer transitions.”
Reduce the effects of quarantine
In early February, news of the virus’s disturbing infection rate began to make headlines, with Barbastathis proposing a project for students in class 2.168. At the end of each semester, class students are tasked with developing a physical model for a problem in the real world and developing machine learning algorithms to address it. He proposed that a team of students work on mapping the spread, which was then known as the “coronavirus”.
The students jumped at the opportunity to work on coronoviruses, which immediately wanted to tackle an occasional problem in a typical MIT fashion, ”says Barbestathis.
Dandekar was one of those students. “The project really interested me because I found this new field of scientific machine learning to deal with a very pressing problem,” he says.
As the world began to spread, the scope of the project grew. Originally started as a project spread within Wuhan, China also grew to spread to Italy, South Korea and the United States.
Both began modeling the spread of the virus in each of these four regions after the 500th case was reported. That milestone marked an apparent confusion as to how various governments implemented quarantine orders.
Armed with accurate data from each of these countries, the research team took the standard SEIR model and augmented it with a neural network that learns how infected individuals under quarantine affect the rate of infection. They trained neural networks through 500 iterations to teach themselves how to predict patterns in the spread of infection.
Using this model, the research team was able to make a direct connection between quarantine measures and effective reproductive number reduction of the virus.
“The neural network is learning what we are calling ine quarantine control power functions,” explains Dandekar. In South Korea, where strong measures were quickly implemented, the quarantine control power function has been effective in reducing the number of new infections. In the United States, where quarantine measures have been slowly rolled out since mid-March, it has been more difficult to stop the spread of the virus.
As the number of cases in a particular country decreases, the forecasting model changes from an exponential rule to a linear one. Italy began entering this linear regime in early April, along with the U.S. Also did not lag behind.
The machine learning algorithm Dandekar and Barbastathis predicted that the United States would begin to move from an exponential regime to a linear regime in the first week of April, with deadlines in the infected case likely to occur on April 15 and April 20. It also states that infection numbers in the United States will reach 600,000 before the rate of infection stabilizes.
“This is a very important moment in time. If we relax the quarantine measures, it can cause disaster, ”says Barbastathis.
According to Barbathasis, one has to look at Singapore only to see the dangers that can prevent quarantine measures very quickly.