Influenza Activity and Province-level Weather Variations in Thailand, 2009 to 2014, Using Random Forest Time-series Approach

Abstract

Influenza activity in Thailand, like many tropical regions, is a burden on public health and economy. The goal of this study is to find links between influenza activity and weather variations of the province-level in Thailand. We use a Random Forest time-series model to analyze the data and evaluate the importance of each variable. Data from 2009 to 2013, in weekly units, is used as a training set to create prediction models. Data from 2014 is used as a test set to validate prediction models. Five high populated provinces are selected to represent different geological regions in Thailand for this study. Although, results indicate that the number of influenza cases from the previous week yields the highest importance for influenza activity, weather variations are not without their impacts. Influenza activity in different provinces is associated with different sets of weather variations and their importance. In all selected provinces, the change in temperatures plays a significant role in influenza activity but its impacts are location-dependent. Correlation coefficients between predicted and observed influenza cases in 2014 are from 0.55 to 0.91.