Disease Spatio-Temporal Analytics Using Attention Based Graph Networks

Abstract

Cholera is an infectious disease caused by ‘Vibro cholerae’ bacteria which is mostly present in warm water and transmitted through drinking water and food which are affected with these bacteria. The growth of the bacteria is affected by environmental factors like temperature, humidity and rainfall and also by the movement of the people. Early studies predicted the outbreaks of cholera in different regions and countries but most of the studies worked on the temporal behavior of the cholera disease. Some studies worked on both spatial and temporal behavior of the disease, but these studies used environmental variables that are sometimes not available for analysis. In this study, we developed a model that will predict the spatial and temporal behavior of the disease by utilizing the advanced deep-learning methods. For that purpose, we used the cholera dataset of Punjab, Pakistan from 2015 to 2018 and developed a Attention based Spatio-temporal Graph Convolutional network model (ASTGCN) as the model provided promising results in recent studies. We trained the model on the given data of cholera to predict the number of patients in each district of Punjab and got the mean square error of 2.2 and compared this result to the with one deep learning model LSTM as the LSTM model showed the mean square error of 3.3. Hence, the ASTGCN model provided better results for the prediction of a cholera outbreak in district of Punjab, Pakistan.