COMBINATION OF UNIVARIATE LONG-SHORT TERM MEMORY NETWORK AND WAVELET TRANSFORM FOR PREDICTING DENGUE CASE DENSITY IN THE NATIONAL CAPITAL REGION, THE PHILIPPINES

DEEP LEARNING AND DWT FOR DENGUE OUTBREAK DETECTION

Abstract

Use of machine learning algorithms on public health big data has paved the way in helping to understand complex associations between diseases and environment. However, in cases where informative prediction of a pathogen spread or transmission is still equivocal, a rapidly deployable independent mathematical modeling method, which does not require a priori knowledge of the pathogen characteristics or requires input parameters predictive of its transmission is needed. This study shows long-short term memory (LSTM) network with wavelet transform has application as a modeling method for dengue incidence. The information derived from this modeling technique can be used to help guide control and prevention measures during an early stage of a dengue surge. Subsequent application of discrete wavelength transform (DWT) to LSTM output resulted in visualization of dengue surges, the majority of which were observed during the wet season (June-December). Augmenting existing disease surveillance systems with automated mathematical models can help focus intervention programs especially in settings where health care resources and surveillance infrastructure are limited. This modeling method has potential applications to other infectious diseases in a developing country setting.