Fundamentals
Bias-variance tradeoff
The bias-variance tradeoff balances systematic underfitting against sensitivity to training data.
How it works
A model that is too rigid misses real structure, while one that is too flexible can fit sample noise. Validation performance reveals the useful middle region.
A concrete example
Illustrative: degree 1 underfits 40 points, degree 14 fits training noise, and degree 4 gives the lowest validation error.
Do not confuse it with Fairness bias
statistical bias is systematic estimation error, not social or demographic unfairness.