Long-term predictors of dengue outbreaks in Bangladesh: A data mining approach

The effects of weather variables on the transmission of vector-borne diseases are complex. Relationships can be non-linear, specific to particular geographic locations, and involve long lag times between predictors and outbreaks of disease. This study expands the geographical and temporal range of previous studies in Bangladesh of the mosquito-transmitted viral infection dengue, a major threat to human public health in tropical and subtropical regions worldwide. The analysis incorporates new compound variables such as anomalous events, running averages, consecutive days of particular weather characteristics, seasonal variables based on the traditional Bangla six-season annual calendar, and lag times of up to one year in predicting either the existence or the magnitude of each dengue epidemic. The study takes a novel, comprehensive data mining approach to show that different variables optimally predict the occurrence and extent of an outbreak. The best predictors of an outbreak are the number of rainy days in the preceding two months and the average daily minimum temperature one month prior to the outbreak, while the best predictor of the number of clinical cases is the average humidity six months prior to the month of outbreak. The magnitude of relationships between humidity 6, 7 and 8 months prior to the outbreak suggests the relationship is multifactorial, not due solely to the cyclical nature of prevailing weather conditions but likely due also to the immunocompetence of human hosts. (C) 2018 The Authors. Production and hosting by Elsevier B.V. on behalf of KeAi Communications Co., Ltd.