Israeli Engineer Invents Algorithm to Predict Worldwide Spread of Coronavirus

Foretell what is yet to happen, That we may know that you are gods! Do anything, good or bad, That we may be awed and see. Isaiah 41:23 (The Israel Bible™)

Israeli Prime Minister Benjamin Netanyahu shocked Israelis recently when he was quoted as saying that by the end of April, a “million Israelis could be infected with Covid-19 and 10,000 will die.” His office did not deny the statement. Prof. Gabi Barabash, a veteran physician, former director-general of the Health Ministry and former director of Tel Aviv Sourasky Medical Center commented on TV that this prediction was “nonsense” and “highly exaggerated.”

How can one accurately predict how fast the coronavirus will spread in Israel and in other countries? 

In a recent paper published in Towards Data Science, a master’s-degree student in electrical engineering named Ran Kremer at Bar-Ilan University in Ramat Gan (near Tel Aviv) explained how an algorithm named “Kalman filter” can accurately predict the worldwide spread of coronavirus and produce updated predictions based on reported data. 

Kalman filter provides estimates of some unknown variables given a series of measurements observed over time. It was pioneered by Rudolf Emil Kalman exactly 60 years ago when it was originally designed and developed to solve the navigation problem in the Apollo space project.  Since then, it has had numerous applications in technology – 

such as guidance, navigation and control of vehicles, computer vision’s object tracking, trajectory optimization, time series analysis in signal processing, econometrics and more.

Kremer recently used Kalman filter to track the spread of coronavirus around the world so as to generate next-day predictions for number of infections, fatalities and recoveries in many infected regions.  The method has produced powerful short-term predictions.  In many regions in China, such as Shanghai, Henan, Beijing and Hubei, as well as for patients on the Diamond Princess cruise ship, Kremer’s predictions on were virtually identical to reality.  Similarly, his model accurately predicted that the spread of coronavirus in China would end in mid-March, with 3,100 deaths in Hubei province.

“With the encouragement of Prof. Zvi Lotker, who gave me the idea to predict the spread of coronavirus before it became the global issue that it is today, I built an online Kalman filter algorithm,” Kremer recalled.  It is adaptive, meaning that it doesn’t need a lot of historical data.  Each day the algorithm is updated with new observations, and after parameter estimation is done it can generate predictions for the next day,” Kremer continued. 

While the Kalman prediction in the short term is very accurate, long-term predictions are more challenging, Kremer said. For long-term predictions, he fit a linear model whose main features are Kalman predictors, infected rate relative to population, time-dependent features and meteorological history and forecasts.  Long-term prediction does not guarantee full accuracy, but it does provide a fair estimation following recent trends.  Running the model on a daily basis can provide better results, he declared.

The engineering student compared COVID-19 to an older fatal virus –Ebola – which is not a new disease (the first cases were identified in 1976) but in 2014 and 2018 erupted again. The fatality rate of Ebola is much higher and may reach a 75% death case comparing to the 3.9% death of COVID-19. As the new coronavirus is an ongoing disease, the fatality rate is not final and will most likely increase, Kremer wrote. The locations and countries are obviously different where Ebola harms mostly in Africa and COVID-19 in China and Asia. Coronavirus seems most likely to spread more in cold weather, unlike Ebola, which spreads more in warm weather. 

In a follow-up article. Kremer – who previously worked for many years in high-tech specializing in data science and machine learning –discussed trends in other hard-hit countries in Asia, Europe and the US and made accurate predictions there as well. Long-term predictions estimate the future trend well but cannot predict when this trend will change, especially during rapid eruptions – as seen in Italy, Spain, Iran and other badly hit countries. This depends on government intervention, local health care and testing capacity. 

The trend of massive eruptions in many European countries and US states (especially New York) keeps growing, although Kremer’s model has begun to predict a downward trend in new infections in Italy, Norway and Sweden starting in mid-April.  In South Korea, the model accurately predicted significant improvement on March 2, 2020.  In Israel, although the number of confirmed cases has recently grown, there is no rapid eruption, and predictions show the spread is significantly lower than in other regions. 

Significant government intervention, alongside tremendous worldwide medical efforts and upcoming warm weather, hopes Kremer, could limit the pandemic soon.  Whenever this happens, this model will rapidly identify it.