Effect of Data from Neighbouring Regions to Forecast Dengue Incidences in Different Regions of Philippines Using Artificial Neural Networks

Abstract

Disease outbreaks forecasting is a vital component in public-health resource planning and emergency preparedness. However, the existing procedures have limitations due to the lack of analytical knowledge about spatiotemporal data. In this paper, we investigate how spatiotemporal data can be leveraged to forecast future disease outbreaks, using dengue incidences in the Philippines for demonstration purposes. Our approach is based on identifying highly correlated regions and using inputs from these regions to train and forecast dengue incidences using Artificial Neural Networks. We then removed the spatial aspect, focusing separately on each region to measure the effect of introducing spatial data. In all the experiments, monthly dengue incidences in 2016 were used as the testing data. Our empirical results show that including spatial data reduces the Mean Absolute Error by approx. 54 % compared to only using data from the target region. We conclude that adding data from neighbouring regions for forecasting can enhance the traditional approaches for forecasting dengue outbreaks, and we recommend that a spatio-temporal analysis is introduced as a standard component of disease outbreak forecasting.