Measles outbreak detection in Metro Manila: comparisons between ARIMA and INAR models

It is the goal of many developing countries to stop the spread of diseases. Part of this effort is to conduct ongoing surveillance of disease transmission to foresee future epidemics. However, in the Philippines, there is a lack of an automated method in
determining their presence. This paper presents a comparison between an integer-valued autoregressive (INAR) model and
the more commonly known autoregressive integrated moving average(ARIMA) models in detecting the presence of disease outbreaks. Daily measles reports spanning from January 1, 2010 to January 14, 2015 were obtained from the Department of Health and were used to motivate this study. Synthetic datasets were generated using a modified Serfling model. Similarity tests using a dynamic time warping algorithm were conducted to ensure that simulated datasets observe similar behavior with the original set. False positive rates, sensitivity rates, and delay in detection were then evaluated between the two models. The results gathered show that an INAR model performs favorably compared to an ARIMA model, posting higher sensitivity rates, similar lag times, and equivalent false positive rates for three-day signal events.