Question: A classifier is clearly underfitting: its training performance is poor and the model seems too limited. Explain why increasing model capacity can help, what risk this change introduces, and how that risk is commonly handled.

Sample answer: Increasing model capacity, such as by adding more hidden units, layers, or tree depth, gives the learner more flexibility to fit the training data. That often lowers avoidable bias when the original model was too small. The trade-off is that a more flexible model can also fit noise in the training set more easily, which raises the chance of overfitting and increases variance. This side effect is usually controlled with regularization, and in some settings with more data or early stopping as well.

Key points:

• Larger models can reduce underfitting and lower avoidable bias.
• Extra capacity helps the model fit the training set better.
• A larger model can also increase variance and overfitting risk.
• Regularization is a common way to limit that added variance.

Rubric: Full credit requires explaining why increasing capacity can improve fit, identifying the possible increase in variance or overfitting, and stating that regularization is used to manage that risk.