Towards an Early Warning System to Combat Dengue

Abstract

Prediction of dengue outbreaks is a timely need for Sri Lanka due to the dramatic increase of dengue incidences. In this thesis we introduce an Early Warning System to predict dengue outbreaks. The proposed model uses the relation between dengue outbreaks with external factors and a neural network based prediction model for outbreak prediction. A simulation model is created which identi es the risk factors, predicts the future outbreaks and the in- tensity of the disease. Proposed model uses epidemiological, metrological and census data for predictions. Data from 2010 to 2012 has been collected for Colombo and Matara districts. It has been identi- ed that there is an inverse relation between rainfall and dengue incidences in Matara district and higher temperature values increase the number of outbreaks. But in Colombo district external factors does not show a direct correlation with dengue cases. A two year data set has been used for training the model while a one year data set is used for testing purposes. The overall training accuracy is 83.6% and 89% for Matara and Colombo districts respectively. The testing accuracy is 62% for Matara and 67% for Colombo districts.