What Is the Bias-Variance Tradeoff?

Bias-Variance Tradeoff Explainer

The bias-variance tradeoff describes two sources of prediction error. High bias means a model makes overly restrictive assumptions and misses real structure, while high variance means its predictions change too much when the training sample changes.

Increasing flexibility can reduce bias because the model can represent more complicated relationships. The same flexibility can raise variance by allowing the model to fit accidents, noise, or quirks in one particular sample.

The useful balance depends on data volume, noise, features, model family, and the cost of different errors. Cross-validation and learning curves estimate how choices behave away from the examples used to fit them.

Bias and variance are diagnostic ideas rather than two dials that can always be measured separately in production. Teams still need error analysis, uncertainty estimates, and monitoring because distribution changes can create failures that the classic tradeoff does not capture.

The bias-variance tradeoff describes the balance between systematic model error and sensitivity to changes in training data.

Key terms include bias, variance, model complexity, training sample, noise, generalization, cross-validation and irreducible error.

High-bias models miss durable structure and tend to underfit.

High-variance models react strongly to sampling noise and tend to overfit.

Validation across several data splits helps compare complexity without judging a model only on its training examples.

The best balance depends on the application and can change when data quality or operating conditions change.

Explore more "Explainers"

Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.