In section 5, we will delve deeper into the topic of overfitting, and why it occurs. First, we cover how to build nonlinear models using linear techniques. This will illustrate how to vary the complexity or flexibility of models. This raises the question: how flexible should our model be? We will tackle this question using the theory of the bias variance decomposition. We will give several practical tools, such as complexity curves and feature curves, to help you tune the complexity of your model. This relates closely to the question: how much data is needed for learning? This question we tackle  using the practical tool of learning curves.

After this section you can:

• • Recognize the concepts bias, variance, irreducible error, overfitting, underfitting
  • Apply learning curves, complexity curves and feature curves
  • Give four options to reduce overfitting and two to reduce underfitting