An online system receives information describing actions performed by online system users, in which the information describes an actual physical location at which each action was performed. The online system retrieves user attributes associated with each user and trains a machine-learning model to predict a physical location of a user based on the information received at the online system and the user attributes. The online system receives information describing a geographic location (e.g., GPS coordinates) from a client device associated with a viewing user of the online system and identifies candidate physical locations within a threshold distance of the geographic location. The online system then uses the trained model to predict a physical location of the viewing user from the candidate physical locations based on user attributes associated with the viewing user and presents content to the viewing user based on the predicted physical location.