Supervised Machine Learning with Python

Error due to variance

In contrast, for the high bias that you're now familiar with, error due to variance can be thought of as the variability of a model's prediction for a given sample. Imagine you repeat the modeling process many times; the variance is how much the predictions for a given sample will vary across different inductions of the model. High variance models are commonly referred to as overfitting, and suffer the exact inverse of high bias. That is, they do not generalize enough. High variance usually comes from a model's insensitivity to the signal as a result of its hypersensitivity to noise. Generally, as model complexity increases, variance becomes our primary concern. Notice in the diagram that a polynomial term has led to ...