Application of remote sensing data for dengue outbreak estimation using Bayesian network

Abstract

Dengue is an epidemic that is a major endemic health problem found in many countries in tropical and worm area. Weather conditions are important factors directly influence the degree of dengue outbreak. Current practice in the dengue outbreak forecasting relies on the meteorological reports. This study shows an alternative way to estimate dengue outbreak level by using Bayesian network (BN). We employ the satellite based remote sensing data to generate the probability model that can estimate dengue outbreak level. We use publicly available satellite based remote sensing data from the NOAA STAR. The data consist of weekly SMN, SMT, VCI, VHI, TCI indexes as factors for estimating the dengue outbreak in the northeast region of Thailand. In this study, 3 BN models had been generated using expert knowledge, greedy thick thinning algorithm, and combination of expert and greedy thick thinning algorithm. All 3 models are validated with the 10-fold cross-validation and ROC Analysis. The experimental results on real data show that the model automatically generated by greedy tick tinning algorithm performs well on overall estimation of dengue outbreak levels. But for an abnormal situation that the outbreak level is significantly higher than usual, the BN model with combination of expert and greedy thick thinning algorithm perform the best in such situation.