Machine Learning for Risk Prediction of Highly Pathogenic Avian In uenza in the Republic of Korea

There have been 7 outbreaks of highly pathogenic avian influenza (HPAI) in the Republic of Korea since 2003 resulting a serious economic burden on the poultry industry. Due to uncertainty of transmission from migratory birds, which is known as the main source of infection, and transmission between poultry farms linked by livestock-related vehicles, it is very difficult to predict and respond to the epidemic. In this work we aim at forecasting spatio-temporal pattern of HPAI occurrence and identifying risk factors with a machine learning technique based on Random Forest regression. Historical data on HPAI outbreaks in 250 regions from 2014 to 2017 are used as a target. Three types of features are used to train the model: epidemiological features related to information on farms infected in the past, demographic features including the number (density) of farms regarding breeding species (chicken and duck) in an area, and geographical features including the habitats of migratory birds and slaughterhouses. The model provides a highly accurate prediction of both temporal and spatial patterns of HPAI outbreaks. Furthermore, we investigate feature importance to explain which features contribute most to the local outbreak of HPAI. Results show that epidemiological features mainly contribute to prediction of the temporal pattern, while the demographic and environmental features mainly contribute to prediction of the spatial distribution.