Wikipedia has a list of the presidents who failed to win reelection. This includes ten United States presidents.
From the list of US presidents, we can find twenty-two presidents who won re-election (counting Franklin Delano Roosevelt three times).
Total, that's thirty-two attempts with twenty-two successes. That's 68.75%.
The Washington Examiner did a study starting from 1900 through 2004. There were fourteen successes and five failures. That's 73.68%. But since then, Barack Obama won a fifteenth time. That's 75%.
These include presidents like Teddy Roosevelt and Calvin Coolidge who won election as an incumbent after replacing a previous president. And this does not include Lyndon B. Johnson in 1968, when he dropped out (anticipating a loss if he ran).
Past performance does not guarantee future results. But the basic chance is around two-thirds to three-quarters. There are models of these things. However, there aren't really enough data points to make for good models.
There are several models of this. Here's an overview by the Brookings Institution. Warning, they have a distinct liberal bias in general, so read with skepticism.
One problem is that these models tend to use information that they don't have yet. For example, from that overview:
For example, Alan Abramowitz’s “Time for Change” forecasting model, based on the incumbent president’s net approval rating at midyear in the Gallup Poll, the growth rate of real GDP in the second quarter of the election year, and whether the incumbent president’s party has held the White House for more than one term, produced the most accurate prediction of the 2012 presidential election among this set of forecasting models.
- The Republicans have not held the presidency for more than one term. This is the only one. Known.
- Midyear approval rating in the Gallup Poll. Unknown until July 2020.
- The growth rate of real GDP in the second quarter of 2020. Unknown until at least July 2020 and subject to adjustment after that.
So we don't yet know two of the three things that the model uses. We could guess, but then the model would be subject not just to its error but to the error in the guesses.