The plots represent two models. A and B, being fit to the same two data sets, training and validation. Model A is 90.5% accurate at distinguishing blue from red on the training data and 75.5% accurate

Question

Options

• AModel A. It is more complex with a higher accuracy than model B on training data.
• BModel A. It performs better on the boundary for the training data.
• CModel B. It is more complex with a higher accuracy than model A on validation data.
• DModel B. It is simpler with a higher accuracy than model A on validation data.

• A
  6% (2)
• B
  25% (9)
• C
  17% (6)
• D
  53% (19)

Explanation

Model B should be selected because it generalizes better to unseen data - the true goal of any classification model. Its validation accuracy (78.3%) exceeds Model A's (75.5%), and its smaller gap between training and validation accuracy (4.7 percentage points vs. 15 percentage points for Model A) indicates that Model A is overfitting: it memorized the training data rather than learning generalizable patterns.

Why the distractors are wrong:

• A is wrong because higher training accuracy alone is not a reason to select a model - overfitting produces high training accuracy but poor real-world performance.
• B is wrong for the same reason; performing well on the training boundary is a symptom of overfitting, not a virtue.
• C is wrong because Model B is the simpler model (lower training accuracy, smaller overfit gap), not the more complex one - complexity and validation accuracy are conflated here.

Memory tip: Think of training accuracy as a student's performance on practice problems they've already seen, and validation accuracy as performance on the real exam. A student who aced practice but struggles on the real test has overfitted - and you wouldn't trust that student over one with more consistent scores across both.

No community discussion yet for this question.