Big Data/Machine Learning/AI
The contribution of global climatic patterns to forecasting dengue cases in Colombia Adrien Saucier* Adrien Saucier Katia Charland Lais Picinini Freitas Jan Saynisch-Wagner Ying Liu Bertha Restrepo Cesar Garcia Gloria Jaramillo Mabel Carabali Kate Zinszer
Dengue incidence is widely associated with global weather patterns. Certain global climate dynamics, such as El Niño Southern oscillation, influences the living conditions of the disease vectors (Aedes aegypti and Aedes albopictus mosquitoes), which in turn, influences dengue incidence. These climatic patterns could provide valuable lead time in dengue forecasting and enhance early warning systems. The objective of our study was to quantify the contributions of El Niño Southern oscillation and Indian Ocean sea surface temperature in dengue forecasting models for Colombia. For this work, we compared different machine algorithms including support vector machine and extreme gradient boosting machine, and selected the best performing algorithms based on different metrics. Using Shapley Additive explanations method, we measured predictor importance at various lead time horizons. Our preliminary results demonstrate the importance of El Niño Southern oscillation (ENSO) and Indian Ocean sea surface temperatures (IOBW) in predicting dengue incidence, with IOBW outperforming ENSO for longer terms forecasts. Longer term forecasting (-12 and -16 weeks ahead) for all locations varied between a mean scaled absolute error of 0.7 and 1.1, indicating good forecasting potential. Early warning systems could benefit from the consideration of IOSS and ENSO in dengue forecasting to improve lead time for outbreak response efforts.