The rapid spread of COVID-19 infection has prompted a wide range of responses from governments, with the intention to either suppress or mitigate the transmission by using non-pharmaceutical interventions. To forecast the impact of different interventions on the COVID-19 epidemic, conventional micro-simulation models, which rely heavily on parametric methods with pre-determined predictors, have been adopted. However, applying this kind of modelling strategy is challenging because: 1) governments exhibited substantial variations in their adopted measures and these measures have been constantly adjusted as the situation has evolved; 2) people may respond to government policies differently due to their cultural and religious background, expectations and norms, and the influence of both traditional news media and social media; and 3) detailed data on population density and travel patterns may be inaccurate or unavailable. To account for complicated interactions of government responses and the pre-existing policy environment, the team explored the usage of state-of-the-art machine learning models in predicting COVID-19 case numbers using standardised policy indicators from the Oxford COVID-19 Government Response Tracker (OxCGRT). In this talk, Dr Luo will report the recent findings in the performance of several machine learning models, including XGBoost, Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRU). She will also discuss the opportunities and challenges of using machine learning in epidemiology and health outcomes research.