Phase 3 · Machine learning · Evaluation
Bias and variance
The two ways a model can be wrong.
Open this lesson Create a free account
Why it matters
When a model disappoints, the next move depends on why it's wrong. Too simple, and more data won't help; too flexible, and a bigger model makes things worse. Bias and variance are the two names for those failures. Once you can tell them apart from a learning curve, you stop guessing at fixes, whether the model is a churn classifier or a fine-tuned LLM.
What you'll be able to do
- Define bias and variance as two separate sources of prediction error
- Measure both by refitting a model on many training samples
- Read a learning curve to tell a high-bias model from a high-variance one
- Match each fix (more data, regularization, ensembles, bigger models) to the error it reduces
Before this lesson
Next in this phase: Precision, recall and F1