Predicting hospital admissions with integer-valued time series

Prediction of seasonal epidemics have beenwidely treated in the medical literature, with various methods to forecast the future cases of a given disease, when it’s about infectious diseases: compartmental methods that forecast the number of persons at each state of the epidemic (susceptible, infected, resistant) are used(1), as well as methods based on time series (ARMA,ARCH,...)(2), this last method can be successful with large sample of data, assuming their normal distribution. Here, we propose to test time series for count data when the continuous time series are not adapted to predict health activity. The aim of this work is to predict the number of hospital admissions of future weeksat local scale, using methodsfor integer-valued time series:INAR(p) and INGARCH(p,q) methods(3–5), wehavealso test the autoregressive methods for continuous data.