A Robust and Non-parametric Model for Prediction of Dengue Incidence
- PDF / 938,558 Bytes
- 7 Pages / 595.276 x 790.866 pts Page_size
- 99 Downloads / 158 Views
REVIEW ARTICLE
© Indian Institute of Science 2020.
A Robust and Non‑parametric Model for Prediction of Dengue Incidence Atlanta Chakraborty1* and Vijay Chandru2 Abstract | Disease surveillance is essential not only for the prior detec‑ tion of outbreaks, but also for monitoring trends of the disease in the long run. In this paper, we aim to build a tactical model for the surveillance of dengue, in particular. Most existing models for dengue prediction exploit its known relationships between climate and socio-demographic factors with the incidence counts; however, they are not flexible enough to cap‑ ture the steep and sudden rise and fall of the incidence counts. This has been the motivation for the methodology used in our paper. We build a non-parametric, flexible, Gaussian process (GP) regression model that relies on past dengue incidence counts and climate covariates, and show that the GP model performs accurately, in comparison with the other existing methodologies, thus proving to be a good tactical and robust model for health authorities to plan their course of action. Keywords: Epidemic, Dengue, Non-parametric, Gaussian process, Covariance, Kernel, Robust, Tactical model
1 Introduction Dengue is a fast emerging pandemic-prone viral disease transmitted by Aedes aegypti and Aedes albopictus mosquitos. According to the World Health Organisation (WHO), each year, an estimated 390 million dengue infections occur all around the world. Cases across the Americas, South-East Asia and Western Pacific exceeded 1.2 million in 2008 and over 3.2 million in 2 01523. Several precautionary measures include vector control tools, like controlling mosquito populations; however, implementation is a major challenge and effective dengue prevention is rarely achieved, specially in developing countries. Often, it is the emergency vector control operation that is usually applied when an outbreak occurs, such as insecticide fogging. Accurate forecasts of incidence cases, or infected individuals are key to planning and resource allocation of dengue vaccines, medical centres, etc. Previous attempts to model dengue have made use of relatively simple models, such as generalised linear model and ARIMA, exploiting the relationship with other environmental variables4, 7, 14. However, most of the times,
J. Indian Inst. Sci. | VOL xxx:x | xxx–xxx 2020 | journal.iisc.ernet.in
disease dynamics are not well understood and such models may fail to capture that 11. Dengue is closely related to the seasonal changes, rainfall and humidity. Our model is trained on historical incidence data, mean surface temperature, humidity and rainfall, and makes use of a Bayesian non-parametric modelling framework, Gaussian processes (GP) that allows for flexibility in the model, thus being able to forecast the sudden peak increase of the incidence counts. 2 Related Work A study and systematic review of existing dengue modelling methods, conducted by Louis et al.11, has been instrumental in providing us with an overview of current modelling efforts and their
Data Loading...
