Remote sensing based modeling of dengue outbreak with regression and binning classification

We present in this paper the application of data mining techniques using remote sensing data to help modeling the prominent characteristics of dengue and severe dengue outbreak in the northeast of Thailand. Dengue is a lethal mosquito-borne viral disease that has long been a serious public health problem in many tropical and some sub-tropical countries. We investigate the rainfall and other remote sensing variables to correlate their contributions to the 2006-2015 dengue outbreaks in Nakhon Ratchasima province through the regression analysis. Classical and severe dengue cases during the past ten years are then discretized with the two binning methods (equal-width 4 bins and standard deviation based 3 bins) and used as the target for model building. Models are built as the classification and regression tree using CART and C5 algorithms. With the cross validation evaluation method, the models built from the equal-width 4 bins show correct classification performance in the range of 78-86%, whereas the models built from the standard deviation based 3 bins yield the 91-95% accuracy.