Predicting vaccine hesitancy from area-level indicators: A machine learning approach