Abstract
COVID-19 outbreak has become a global pandemic that affected more than 200 countries worldwide. Predicting the behavior of this outbreak has a crucial role in organizing preventive and protective actions, and in improving the decision making process.The aim of predicting the number of people who contract the virus has so far been pursued with regression models (exponential, logistics, ...), but regressions can integrate the variable context only a posteriori. The regression models are all dependent on their own history, thus, they can not display anything which did not happen before.The pandemic infection of COVID-19 presents a transmission behavior that is widely changing over time. This is due to the growth of the efficiency in the detection of infected, for the changes in social distancing measures and for the widespread use of individual protection devices.The approach presented in this paper, starting from the definition of simplified risk assessment framework, aims at designing a probabilistic model for the virus transmission and detection, keeping into account this context changes, binding the correct set of variables to them, and at inferring the distribution for the underlying stochastic variables. This is a key to unlock innovative and valuable insights from the current events. The model has been built in Gen, a probabilistic programming system, built at MIT and embedded in Julia.