It's a rather simple process, although quite time-consuming. We do it this way:
By cross-referencing voter registration records, from which we can determine a voter's home address and the address of the voter's polling place; the voter's tax returns and previous voting habits, from which we can determine where the voter works and logically deduce whether the voter will travel from home or work to the voter's polling place. We also collate the number of registered voters who might vote at a given polling place and compare that to the number of poll-workers and voting machines at that polling place. In addition, we phone-sample voters to determine at what time they expect to vote so we know when there will be delays. Of course, we also factor in absentee voters and early voters, whose effective travel distance is 0 and whose wait time is 0.