The Use of Spaceborne and Oceanic Sensors to Model Dengue Incidence in the Outbreak Surveillance System

摘要

This research focuses on the development of a computational data-driven modeling method to be used in the dengue outbreak surveillance system. The outbreak-level forecasting is based on the estimation of dengue fever cases in Thailand using both statistical and data mining techniques. Major statistical techniques used in this research are linear regression and generalized linear model. The data mining algorithms used in our study are chi-squared automatic interaction detection (CHAID), classification and regression tree, artificial neural network, and support vector machine. The input data are from four sources, which are remotely sensed indices from the NOAA satellite to represent vegetation health and other related weather conditions, rainfall, the oceanic Nino index (ONI) for justifying climate variability affecting amount of rainfall, and historical dengue cases in Thailand to be used as the modeling target. In the modeling process, these data are lagged from 1 up to 24 months to observe time-series effect. On comparing performances of models built from different algorithms, we found that CHAID is the best one yielding the least error on estimating dengue cases. From the CHAID models to forecast dengue cases in Bangkok metropolitan and Nakhon Ratchasima in the northeast of Thailand, the high level of ONI is the most important factor. The large amount of rainfall is significant factor contributing to dengue outbreak in Chiang Mai in the north and Songkhla in the south.