Predictive model for the dengue incidences in Sri Lanka using mobile network big data

Abstract

Multiple studies in recent literature have shown that human mobility has a considerable effect on the importation of dengue to an immunologically dengue ‘naive’ regions. The purpose of this study is to validate whether this fact is applicable to dengue endemic regions as well as assess the impact of human mobility when developing predictive models for dengue incidence of a particular geographical region. The information regarding human population mobility of Sri Lanka was derived using Mobile Network Big Data (MNBD) due to its better representativity and accuracy over other conventional data sources such as census data or surveys. The derived mobility values for each region of the country is then weighted using reported past dengue cases of the relevant region. We used Artificial Neural Networks and XGBoost to predict the dengue incidence for selected administrative regions within Western Province of Sri Lanka. Accuracy of the prediction for dengue incidence improved on average when considering all the administrative regions included in our study by incorporating weighted human mobility. Our study shows that there is an impact of human mobility on the spread of Dengue fever in Sri Lanka while introducing a generalizable methodology to fuse big data sources with traditional data sources in developing predictive models using machine learning techniques.