Ground Truth

The verified, correct answer or label used as a reference when evaluating model predictions. In supervised learning, ground truth comes from human-labeled datasets. Its quality directly impacts the reliability of training and evaluation.