Probabilistic prediction of COVID-19 infections for China and Italy, using an ensemble of stochastically-perturbed logistic curves
share
The spread of COVID-19 has put countries under enormous strain, and any method that could give some valuable forecasts of how the number of infected people grow should be exploited. In particular, Italy is now facing an unprecedented number of people in intensive care units, and has taken drastic measures to contain the spread of the disease. An ensemble-based, probabilistic prediction approach is used to predict COVID-19 infection numbers. The forecasts are generated using an ensemble of stochastically-perturbed logistic curves, with parameters estimated by randomly perturbing the observed data, in a way to simulate observation errors. An ensemble of 20 members is used to compute the most likely outcome, ranges and probabilities. To our knowledge, this is a novel; technique for this field. Firstly, we show that the logistic equation is capable to fit very well observed data of COVID-19 for China. Secondly, for China we show that using COVID-19 data for the first 18 days of the spread (22 January to 9 February 2020), we could provide very valuable probabilistic forecasts for the next 27 days (i.e. up to the time of writing, 10 March). Thirdly, we apply the same method to Italy, using logistic curves estimated using data for the first 18 days of the spread of the disease (21 February to 10 March 2020). Ensemble-based probabilistic forecasts indicate that values will continue to rise till day 35 (27th of March). They indicate that on day 45 (6th of April) there is a 70 between 22,700 and 29,000. The most likely value, given by the ensemble-mean forecast, is about 25,800. The comparison of normalized, growth curves for China and Italy shows that during the first 18 days, the doubling time has been slightly faster in China than in Italy, while after day 15 it reached the value of 2.7 days, close to the asymptotic value of 2.8 days.
READ FULL TEXT
