Bias-Variance Tradeoff
Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary
The balance every predictive model strikes between error from oversimplification (bias) and error from fitting noise in the training data (variance).
What is the Bias-Variance Tradeoff?
The bias-variance tradeoff is the balance every predictive model strikes between two sources of error: bias, from a model too simple to capture the pattern, and variance, from a model so flexible it fits the noise. Reducing one tends to raise the other.
Bias here means statistical estimation error, not the discriminatory outcomes that bias refers to in AI fairness discussions. The two concepts share a word and nothing else.
How the Tradeoff Works
A high-bias model imposes assumptions the data does not obey – a straight line forced through a curve. It misses the pattern the same way on every dataset, which is underfitting.
A high-variance model bends to whatever sample it sees. Retrain it on a different draw of the same data and its predictions change substantially, because it fit that sample’s noise, which is overfitting.
Total error is roughly bias plus variance plus irreducible noise, and the minimum sits between the extremes. Model selection, hyperparameter tuning, and regularization are all ways of searching for that point, with cross-validation as the measuring instrument.
Example of the Bias-Variance Tradeoff
Fit three models to the same 100 house sales, predicting price from floor area. A straight line predicts stably but underestimates every large house: high bias. A 15th-degree polynomial passes near every training point, yet swings wildly between them, and retraining on a different 100 sales produces a completely different curve: high variance.
A gently curved model in between misses individual sales by a little but tracks the real area-to-price relationship. On new listings, it beats both extremes. That middle model has not eliminated either error source; it has balanced them where their sum is smallest.
Related AI terms: Overfitting · Underfitting · Regularization · Cross-Validation · Model Evaluation
Did you like the Bias-Variance Tradeoff gist?
Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.
Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About