Machine Learning Basics: Where Bias and Variance Fit in Overfit–Underfit Overfit is a condition that treats noise in training data as a reliable indicator rather than an anomaly.