Question: A fraud classifier has 14% error on the training set and 21% error on the development set. Explain why it is not accurate to describe this model as only underfitting or only overfitting. Include the meaning of its training and development performance in your response.

Sample answer: This model shows signs of both problems at once. The training error is still fairly high, which suggests the model is not fitting the training data well enough and therefore has high bias. The development error is even higher, which suggests the model is also failing to generalize well and has high variance. Because both indicators are present, a single label such as only underfitting or only overfitting does not fully describe the model.

Key points:

• The model shows both underfitting and overfitting symptoms.
• High training error indicates poor fit to the training data and suggests high bias.
• Higher development error shows weaker generalization and suggests high variance.

Rubric: The response must explain that the classifier shows both underfitting and overfitting, state that the training error is high enough to indicate poor fit/high bias, and state that the development error is worse, indicating high variance and weak generalization.