Changepoint Detection Through Segmented Autoregressive Moving Average (ARMA) Models for the Study of Epidemics – A Bayesian Approach

Mr. Russell Latterman
PhD Student
University of Wisconsin – Milwaukee

Changepoint detection often involves the discovery of sudden abrupt and sustained fluctuations in population dynamics over time. This may reflect a change in population parameters, or it may be that we are unknowingly drawing data from separate populations or groups. If we assume data following a changepoint to be independent of previous data, we may analyze data on segments delimited by changepoints. If data within segments are considered correlated, it may be appropriate to use an Autoregressive Moving Average (ARMA) model on each segment. Publications on segmented ARMA modeling are relatively limited, and the application of such a model for the study of epidemics is surprisingly absent from the literature. The challenges that arise when implementing such models will be demonstrated, and potential for applications in epidemiology will be discussed.