Model Evaluation and Validation Flashcards

High training accuracy but low test accuracy is a classic symptom of:

Answer: Overfitting

Overfitting means the model memorizes training data but fails to generalize.

In the bias-variance tradeoff, a model with high bias typically:

Answer: Underfits and performs poorly on both training and test data

High bias means the model is too simple, underfitting both sets.

Which technique most directly reduces overfitting by penalizing large coefficients?

Answer: Regularization (L1/L2)

L1/L2 regularization adds a penalty term that shrinks coefficients, reducing variance.

Early stopping prevents overfitting during training by:

Answer: Halting training when validation performance stops improving

Early stopping ends training once validation loss begins to rise, avoiding over-training.

A learning curve where both training and validation errors remain high suggests:

Answer: Underfitting (high bias)

Persistently high errors on both sets indicate the model is too simple.

Data leakage in model validation typically causes:

Answer: Overly optimistic validation scores that collapse in production

Leakage lets information from the target or test set inflate validation metrics unrealistically.

Increasing model complexity generally has what effect on the bias-variance tradeoff?

Answer: Decreases bias but increases variance

More complex models fit better (lower bias) but become more sensitive to data (higher variance).