Variance measures how much a model’s predictions change when trained on different subsets of the training data. It reflects the model’s sensitivity to the specific data used for training.

High Variance

  • The model is too sensitive to the training data.
  • Captures noise as well as signal; poor generalization.
  • Symptom: Good performance on training data, poor performance on test data (overfitting).

Bias–Variance Context

  • Variance = Error due to instability in the model (changes drastically with small variations in training data).
  • High flexibility; higher variance.
  • Low variance models = more stable but less flexible.

Intuition

If you slightly change the training data and the predictions change a lot; high variance. If predictions remain almost the same; low variance.

Key Point

Variance is about model sensitivity and stability:

Related: