Bias-variance tradeoff quiz

Question 1

What does 'bias' refer to in the bias-variance tradeoff?

Accepted Answer

Error from a model being too simple to capture the true pattern

Answer

Error caused by noisy measurement instruments only

Answer

The total number of parameters in a model

Answer

The speed at which a model is trained

Question 2

What does 'variance' refer to in the bias-variance tradeoff?

Accepted Answer

How much predictions change with different training sets

Answer

The average error across all possible models

Answer

The bias of the data collection process

Answer

The number of features used by the model

Question 3

How are high bias and underfitting related?

Accepted Answer

High bias is the cause of underfitting

Answer

High bias always causes overfitting

Answer

Underfitting only happens with too much data

Answer

They are completely unrelated concepts

Question 4

How are high variance and overfitting related?

Accepted Answer

High variance is the cause of overfitting

Answer

High variance lowers test error as training continues

Answer

High variance always means underfitting

Answer

They have no connection

Question 5

What is the bias-variance tradeoff?

Accepted Answer

Lowering one error type tends to raise the other, so balance is needed

Answer

Bias and variance can both be driven to zero easily

Answer

Variance only matters for classification problems

Answer

Bias and variance always increase together

Question 6

In the standard decomposition, total expected prediction error is made of which three parts?

Accepted Answer

Bias squared + variance + irreducible error

Answer

Bias + variance + accuracy

Answer

Training error + test error + noise

Answer

Mean + median + mode of errors

Question 7

What is irreducible error and why can't a model eliminate it?

Accepted Answer

Noise inherent in the data that no model can remove

Answer

Error from using too few training epochs

Answer

Error that disappears with enough regularization

Answer

The bias of the chosen algorithm

Question 8

As model complexity increases, what typically happens to bias and variance?

Accepted Answer

Bias decreases and variance increases

Answer

Both bias and variance decrease

Answer

Bias increases and variance decreases

Answer

Both stay constant

Question 9

A model has low training error but high test error. Is this high bias or high variance?

Accepted Answer

High variance (overfitting)

Answer

High bias (underfitting)

Answer

Irreducible error only

Answer

Perfectly balanced model

Question 10

A model has high training error and similarly high test error. Is this high bias or high variance?

Accepted Answer

High bias (underfitting)

Answer

High variance (overfitting)

Answer

Just right model

Answer

Only data leakage

Question 11

Why does increasing the degree of a polynomial regression eventually raise variance?

Accepted Answer

More flexibility lets it fit noise, so predictions vary with the sample

Answer

Higher degree always lowers test error

Answer

More parameters reduce variance

Answer

Polynomials cannot overfit

Question 12

How does regularization (like L2/Ridge) affect bias and variance?

Accepted Answer

It reduces variance while slightly increasing bias

Answer

It reduces both bias and variance to zero

Answer

It increases variance and reduces bias

Answer

It has no effect on either

Question 13

What is the effect of increasing the size of the training dataset on variance?

Accepted Answer

It tends to reduce variance

Answer

It tends to increase variance

Answer

It mainly reduces bias

Answer

It has no effect on test error

Question 14

Why won't simply adding more training data fix a high-bias (underfitting) model?

Accepted Answer

The model is too simple to capture the pattern regardless of data amount

Answer

More data always fixes any error

Answer

High bias is caused by too much data

Answer

Adding data increases bias further

Question 15

How does a learning curve help diagnose bias versus variance?

Accepted Answer

A large gap suggests high variance; high converged error suggests high bias

Answer

A flat line always means perfect fit

Answer

It only measures training time

Answer

Gaps in the curve indicate irreducible error only

Question 16

Which type of error does k-fold cross-validation primarily help estimate and control?

Accepted Answer

Generalization error, helping balance bias and variance

Answer

Only the training error

Answer

Only the irreducible error

Answer

The number of model parameters

Question 17

For a k-nearest-neighbors model, how does increasing k affect bias and variance?

Accepted Answer

Larger k increases bias and decreases variance

Answer

Larger k decreases both bias and variance

Answer

Larger k decreases bias and increases variance

Answer

k has no effect on bias or variance

Question 18

How does the depth of a decision tree relate to bias and variance?

Accepted Answer

Deeper trees lower bias but raise variance

Answer

Deeper trees raise bias and lower variance

Answer

Tree depth affects neither

Answer

Shallow trees always overfit

Question 19

Why does bagging (e.g., Random Forest) reduce variance?

Accepted Answer

Averaging many models cancels random fluctuations

Answer

It increases the depth of a single tree

Answer

It removes the irreducible error

Answer

It mainly reduces bias not variance

Question 20

Boosting primarily targets which component of error, and how?

Accepted Answer

It mainly reduces bias by sequentially correcting errors

Answer

It mainly reduces variance by averaging

Answer

It eliminates irreducible error

Answer

It has no effect on bias

Question 21

Why is the most complex model not always the best choice?

Accepted Answer

Its high variance can hurt generalization despite low bias

Answer

Complex models always have high bias

Answer

Complex models cannot fit training data

Answer

It always has the lowest test error

Question 22

In the formula Error = Bias^2 + Variance + Irreducible, why is bias squared rather than just bias?

Accepted Answer

It arises from expanding the expected squared error

Answer

Bias is always negative so must be squared

Answer

Squaring removes the variance term

Answer

It is an arbitrary convention with no reason

Question 23

If a model's predictions are consistently off in the same direction regardless of the data sample, which error dominates?

Accepted Answer

High bias

Answer

High variance

Answer

Pure irreducible error

Answer

Zero error

Question 24

Using a dartboard analogy, what do tightly clustered darts far from the bullseye represent?

Accepted Answer

High bias, low variance

Answer

Low bias, high variance

Answer

Low bias, low variance

Answer

High bias, high variance

Question 25

Using the dartboard analogy, what do darts scattered widely but centered around the bullseye represent?

Accepted Answer

Low bias, high variance

Answer

High bias, low variance

Answer

Low bias, low variance

Answer

High bias, high variance

Question 26

How does early stopping in neural network training relate to the bias-variance tradeoff?

Accepted Answer

It stops training before variance grows too large

Answer

It increases variance to reduce bias

Answer

It removes irreducible error

Answer

It has no link to overfitting

Question 27

Which of these is a valid strategy to reduce variance without changing the model family?

Accepted Answer

Add more training data or apply regularization

Answer

Remove all features from the model

Answer

Train on a smaller dataset

Answer

Increase the polynomial degree

Question 28

What does it mean for a model to 'generalize well' in terms of bias and variance?

Accepted Answer

It keeps both bias and variance low enough for low test error

Answer

It memorizes the training data perfectly

Answer

It only minimizes training error

Answer

It maximizes the number of parameters

Question 29

Why does adding more irrelevant features tend to increase variance?

Accepted Answer

Extra freedom lets the model fit noise, raising variance

Answer

Irrelevant features always reduce variance

Answer

More features only reduce bias with no cost

Answer

Features have no effect on variance

Question 30

A teacher fits a degree-1 line and a degree-20 curve to the same noisy data. Which has higher bias and which has higher variance?

Accepted Answer

Degree-1 has higher bias; degree-20 has higher variance

Answer

Degree-1 has higher variance; degree-20 has higher bias

Answer

Both have identical bias and variance

Answer

Degree-20 has both lower bias and lower variance with no cost

Test yourself