Accuracy refers to the percentage chance that a given attack will hit. An attack may have high accuracy or low accuracy, but both of these are a measurement of accuracy.
Common usage also applies the word accuracy to include things that we don't normally think of as a "hit", such as a skill check, but it makes more sense if we remember that the DC of a roll is what's called its "target number", as per PHB p.7:
3. Compare the total to the target number. If the total equals or exceeds the target number, the ability check, attack roll or saving throw is a success. Otherwise, it's a failure. The DM is usually the one who determines target numbers and tells players whether their checks, attack rolls, or saving throws succeed or fail.
The target number for an ability check or saving throw is called Difficulty Class (DC). The target number for an attack roll is called an Armor Class (AC).
CTWind's answer cites an informative 2012 article which appears to beis the origin of the term "bounded accuracy", describing it initially in terms of attacks, then extending the concept orthogonally to other d20 rolls. Hence the reason why it's called "accuracy" even when it doesn't refer to attacks is that it's essentially the same mechanic as when it does refer to attacks.
In earlier editions of D&D, particularly third edition, numbers could get very high where high-level or highly-optimal characters with a lot of stacking bonuses were involved. A roll could be almost impossible for one character and almost guaranteed success for another.
Hence bounded accuracy, where the range of numbers it's possible for a character to roll is kept within tighter limits, or bounds. This shifts the focus of the game away from character optimization (by stacking small bonuses), and toward interacting with the fictional world by searching for in-situation ways to acquire Advantage.
