Towards Predicting Dengue Fever Rates Using Convolutional Neural Networks and Street-Level Images

The ability to identify urban locations with a high risk of diseases infections is a central aspect of public policies aiming at controlling these diseases. In this paper, we investigate the use of street-level images, such as those from Google Street View, along with Convolutional Neural Networks to predict Dengue Fever (DF) and Dengue Hemorrhagic Fever (DHF) rates in urban locations. We conduct a case study in the city of Rio De Janeiro, Brazil, using the proposed methodology and DF/DHF data between the years of 2010 to 2014. We compare two Siamesebased CNNs, yielding an overall accuracy of 67% for 20,400 different locations. We conclude that street-level images are useful for the problem.