Dengue outbreak prediction for GIS based Early Warning System

Dengue fever is mostly found in the tropical and sub-tropical regions of the world. In the recent five years, Jakarta is one of the five provinces with the highest Incidence Rate (IR) in Indonesia. To reduce the IR, early detection of dengue fever is an important preventive effort. Therefore, we developed a Dengue Early Warning System (DEWS) to detect the potential of outbreaks of dengue virus based statistical calculations and GIS. The aim of this study is to analyze the performance of DEWS by testing its accuracy of predictions, using data of environmental factors, climate and surveillance in District Cempaka Putih. Naïve Bayes was chosen as Dengue outbreak predictor. Through the process of selecting a subset of attributes (Feature Subset Selection) with exhaustive search approach and Naïve Bayes accuracy as feature subset quality evaluation criteria, as the result we identified four attributes that contributed significantly to the prediction accuracy. The four attributes are house density, free larvae index, container potential nest larvae, and average rainfall in the last 2 months. The system achieved an accuracy of 97.05% in term of Geometric Mean. Further error analysis revealed that the sensitivity, specificity, Positive Predicted Value, and F1 of the system were 94.52%, 99.65%, 98.57% and 96.50%, respectively.