Abstract:Dengue is estimated to have approximately 390 million infections annually out of which an estimated 96 million manifest (Bhatt et al., 2013). WHO estimates almost half the global population to be at risk from this neglected tropical infectious disease. The enormous economic burden of the disease is evident when considering that an estimated 264 disability-adjusted life years (DALYs) per million population is lost due to dengue each year (World Health Organization, 2012). In Sri Lanka, it was estimated that within the Colombo district, where the nation’s capital is situated, a financial burden of US$ 971,360 was imposed upon the national health system in 2012 alone just for the execution of preventive measures (Thalagala et al., 2016). Considering all of these factors, optimizing resource allocation and reducing the economic burden of dengue should be a key element of any long term strategy for dealing with the disease. In this context, the ability to predict dengue outbreaks for a particular region 2 weeks in advance would lead to better resource mobilization and would be invaluable asset for the public health sector. Additionally, a disease outbreak forecasting model developed for dengue need not be limited to tackling resource allocation for that disease only, but can also be used to forecast outbreaks of other arboviral diseases such as Ebola, Zika or Chikungunya. In Sri Lanka at least, Chikungunya is already prevalent and it is just a matter of time before Zika comes to Sri Lanka given that cases have been detected in the Asian region including Singapore. In an epidemic or outbreak of an infectious disease, we would need to develop a deep understanding of human mobility patterns within the infected regions to identify the potential hotspots and also to forecast to which regions are most likely to be infected next. For the purpose of understanding human mobility in disease propagation, Mobile Network Big Data (MNBD) has become a low cost data exhaust that provides rich insight into human mobility patterns with better spatial and temporal granularity when compared to statistical methods which rely mostly on macro level population parameters. In this work, we evaluate multiple machine learning techniques such as Neural Networks (NN), Support Vector Machines (SVM), Random Forests and XGBoost to determine which technique performs best. A comparison of the model performance between different techniques is provided. We go on to use a genetic algorithm based optimization to further improve the accuracy of these models. Our work shows that Call Detail Records (CDR) can be used to derive proxy indicators for human movement patterns which are applicable across multiple machine learning models. Our results show that human mobility has an impact on dengue incidence, even in dengue endemic regions. The forecasting models developed in this work can be utilized to effect a significant impact on the issue of allocating resources effectively to combat dengue which in turn would lead to reduced economic burden as well as reduced mortality and morbidity.

SARIMA Forecasts of Dengue Incidence in Brazil, Mexico, Singapore, Sri Lanka, and Thailand: Model Performance and the Significance of Reporting Delays

Assessing dengue forecasting methods: A comparative study of statistical models and machine learning techniques in Rio de Janeiro, Brazil

Ensemble Approaches for Robust and Generalizable Short-Term Forecasts of Dengue Fever. A retrospective and prospective evaluation study in over 180 locations around the world

Forecasting Dengue Fever in Brazil Using Multimodal Data, Including Climate Data

Infrastructure and methods for real-time predictions of the 2014 dengue fever season in Thailand

A reproducible ensemble machine learning approach to forecast dengue outbreaks

Dengue prediction by the web: Tweets are a useful tool for estimating and forecasting Dengue at country and city level

Improving Dengue Forecasts by Using Geospatial Big Data Analysis in Google Earth Engine and the Historical Dengue Information-Aided Long Short Term Memory Modeling

Improving Disease Outbreak Forecasting Models for Efficient Targeting of Public Health Resources

Developing a Dengue Forecast Model Using Machine Learning: A Case Study in China.

Time Series Methods and Ensemble Models to Nowcast Dengue at the State Level in Brazil

Forecasting dengue and influenza incidences using a sparse representation of Google trends, electronic health records, and time series data

Spatiotemporal analysis of historical records (2001-2012) on dengue fever in Vietnam and development of a statistical model for forecasting risk

Precision Prediction for Dengue Fever in Singapore: A Machine Learning Approach Incorporating Meteorological Data

Interdisciplinary modelling and forecasting of dengue

Faster indicators of dengue fever case counts using Google and Twitter

An open challenge to advance probabilistic forecasting for dengue epidemics

Dengue nowcasting in Brazil by combining official surveillance data and Google Trends information

Leveraging Climate Data for Dengue Forecasting in Ba Ria Vung Tau Province, Vietnam: An Advanced Machine Learning Approach

Prediction of dengue fever outbreaks using climate variability and Markov chain Monte Carlo techniques in a stochastic susceptible-infected-removed model

When climate variables improve the dengue forecasting: a machine learning approach