Forecasting dengue epidemics using a hybrid methodology

Dengue case management is an alarmingly important global health issue. The effective allocation of resources is often difficult due to external and internal factors imposing nonlinear fluctuations in the prevalence of dengue fever. We aimed to construct an early-warning system that could accurately forecast subsequent dengue cases in three dengue endemic regions, namely San Juan, Iquitos, and the Philippines. The problem is solely regarded as a time series forecasting problem ignoring the known epidemiology of dengue fever as well as the other meteorological variables. Autoregressive integrated moving average (ARIMA) model is a popular classical time series model for linear data structures whereas with the advent of neural networks, nonlinear structures in the dataset can be handled. In this paper, we propose a novel hybrid model combining ARIMA and neural network autoregressive (NNAR) model to capture both linearity and nonlinearity in the datasets. The ARIMA model filters out linear tendencies in the data and passes on the residual values to the NNAR model. The proposed hybrid approach is applied to three dengue time-series data sets and is found to give better forecasting accuracy in comparison to the state-of-the-art. The results of this study indicate that dengue cases can be accurately forecasted over a sufficient time period using the proposed hybrid methodology.