Learning from panel data of dengue incidence and meteorological factors in Jakarta, Indonesia

Abstract

Medical statistics collected by WHO indicates that dengue fever is still ravaging developing regions with climates befitting mosquito breeding amidst moderate-to-weak health systems. This work initiates a study over 2009–2017 panel data of dengue incidences and meteorological factors in Jakarta, Indonesia to bear particular understanding. Using a panel random-effect model joined by the pooled estimator, we show positively significant relationships between the incidence level and meteorological factors. We ideate a clustering strategy to decompose the meteorological datasets into several more datasets such that more explanatory variables are present and the zero-inflated problem from the incidence data can be handled properly. The resulting new model gives good agreement with the incidence data accompanied by a high coefficient of determination and normal zero-mean error in the prediction window. A risk measure is characterized from a one-step vector autoregression model relying solely on the incidence data and a threshold incidence level separating the low-risk and high-risk regime. Its magnitude greater than unity and the weak stochastic convergence to the endemic equilibrium mark a persistent cyclicality of the disease in all the five districts in Jakarta. Moreover, all districts are shown to co-vary profoundly positively in terms of epidemics occurrence, both generally and timely. We also show that the peak of incidences propagates almost periodically every year on the districts with the most to the least recurrence: Central, South, West, East, and North Jakarta.