Abstract
The Pacific Island Countries and territories (PICs) experienced a doubling of annual reported dengue outbreaks between 2012 to 2019, including concurrent outbreaks of multiple dengue serotypes. This has major health implications for the region as reinfection can lead to more serious health complications. Decision support systems for dengue can mitigate the risk of outbreaks by providing information on which early planning and proactive interventions may be based. Such decision support systems require an understanding of the factors that drive dengue outbreaks. Current efforts to build decision support tools, such as disease forecasting models, rely on links between environmental factors and dengue outbreaks, largely ignoring human movement. To address this gap we used random forest and XGBoost models to analyse potential links between human movement and meteorological variables on dengue outbreaks in PICs. We used variable importance metrics and a forward selection process to identify key combinations of explanatory variables. The findings highlighted that the two-month lead average minimum temperature was an important indicator of both months when an outbreak was current ("outbreak month") and the month of the start of outbreaks ("start month"). In comparison, international arrivals from outside the Pacific Islands was only considered important for the start month. These results were consistent whether random forest or XGBoost was used to build classifier models. Despite some differences in variables selected, forward selection resulted in similar performance for both random forest and XGBoost models. The models developed in this study were exploratory and require further development before use as a policy tool. Future research into dengue risk in PICs should further explore the impact of human mobility between countries on dengue outbreaks.