Using the time varying Kalman filter for prediction of Covid-19 cases in Latvia and Greece

Abstract

In this work we study applicability of Kalman filters as decision support for early warning and emergency response system for infectious diseases as CoVID-19. Here we use only the actual observations of new cases/deaths from epidemiological survey. We investigated the behavior of various time varying measurement driven models. We implement time varying Kalman filters. Preliminary results from Greece and Latvia showed that Kalman Filters can be used for short term forecasting of Co Vid-19cases. The mean percent absolute error may vary by model; some models give satisfactory results where the mean percent absolute error in new cases is of the order of 2%-5%. The mean absolute error in new deaths is of the order of 1–2 deaths. We propose the use of Kalman Filters for short term forecasting, i.e. next day, which can be a useful tool for improved crisis management at the points of entry to a country or hospitals.