Prediction of Dengue cases in National Capital Region, Philippines using artificial neural networks

Abstract

Dengue cases in the National Capital Region, Philippines were predicted using three-layer artificial neural networks (ANNs) with 4 input neurons, 40 hidden neurons, and 1 output neuron. Available data from the Department of Health and Philippine Atmospheric, Geophysical, and Astronomical Services Administration were used to assess the dependence of dengue cases on the monthly lagged values of cumulative rainfall, mean relative humidity, and mean temperature. The ANNs yield modest 24-month predictions of dengue cases for ANNs that are trained at a goal sum squared error of 0.05. Furthermore, the selection ratio between the training and test set foretells the goodness of the predicted dengue cases. Statistical analyses indicate that, given the proper choice of goal error, ANNs can be used to devise schemes to mitigate dengue outbreaks in the Philippines. While the selection ratio also determines the goodness of dengue case predictions, the underlying behaviour remains to be understood.