Overfitting explained simply quiz

Question 1

What does "overfitting" mean when a computer learns from examples?

Accepted Answer

Learning the training data too well, including its noise, so it fails on new data

Answer

Learning too slowly to be useful

Answer

Using too little data to train

Answer

Having too few features to learn from

Question 2

A student memorizes every answer in last year's exam paper but can't solve new questions. How is this like overfitting?

Accepted Answer

The model memorizes examples instead of learning general rules, so it fails on new data

Answer

The model learns faster than the student

Answer

The model forgets everything it studied

Answer

The model needs more exams to pass

Question 3

What is the opposite problem of overfitting called?

Accepted Answer

Underfitting

Answer

Superfitting

Answer

Overlearning

Answer

Crossfitting

Question 4

Why do we test a model on data it has never seen before?

Accepted Answer

To check if it really learned the pattern instead of just memorizing

Answer

To make training faster

Answer

To use up extra data

Answer

To confuse the model on purpose

Question 5

What is "noise" in data, and why is it a problem for overfitting?

Accepted Answer

Random meaningless variation that an overfit model wrongly treats as a pattern

Answer

Loud sounds recorded near the computer

Answer

The correct answers in the dataset

Answer

The speed of the training process

Question 6

How can you usually tell that a model is overfitting by looking at its scores?

Accepted Answer

High training accuracy but low test accuracy (a big gap)

Answer

Equal low scores on both training and test

Answer

High test accuracy but low training accuracy

Answer

Identical perfect scores everywhere

Question 7

Why does splitting your data into a training set and a test set help with overfitting?

Accepted Answer

It lets you check performance on data the model never trained on

Answer

It makes the dataset twice as big

Answer

It deletes the noise automatically

Answer

It speeds up the computer

Question 8

Does making a model more complex always make it better? Explain.

Accepted Answer

No, too much complexity causes overfitting and hurts new-data performance

Answer

Yes, more complex models are always better

Answer

Yes, complexity removes all errors

Answer

No, complexity has no effect at all

Question 9

How does giving a model more training data help reduce overfitting?

Accepted Answer

More varied examples make the true pattern clearer and memorizing harder

Answer

More data always causes more overfitting

Answer

More data slows learning so it cannot overfit

Answer

Data amount has no effect on overfitting

Question 10

A model recognizes the dogs in its training photos perfectly but fails on new dog photos. What likely happened?

Accepted Answer

It overfitted to specific training photos instead of learning general dog features

Answer

It learned too little to recognize anything

Answer

The new photos are not really dogs

Answer

It needs a faster computer

Question 11

What is "generalization" in machine learning?

Accepted Answer

Performing well on new, unseen data, not just training data

Answer

Making the training time longer

Answer

Memorizing all the training answers

Answer

Using the largest possible model

Question 12

Imagine drawing a line through scattered dots. How does a straight line vs. a super wiggly line relate to fitting?

Accepted Answer

The wiggly line that hits every dot is overfitting; a smooth fit is better

Answer

The wiggly line is always the best choice

Answer

The straight line always overfits

Answer

Both lines fit equally well always

Question 13

What does "regularization" do to help prevent overfitting?

Accepted Answer

It penalizes excessive complexity to keep the model simpler

Answer

It adds more noise to the data

Answer

It makes the model memorize faster

Answer

It deletes the test set

Question 14

Why is memorizing not the same as learning, for both humans and machines?

Accepted Answer

Learning understands the idea for new cases; memorizing only repeats known answers

Answer

They are exactly the same thing

Answer

Memorizing is always better than learning

Answer

Machines cannot memorize at all

Question 15

What is a "validation set" and how does it help catch overfitting early?

Accepted Answer

A held-out set checked during training to spot overfitting early

Answer

The same data used for training

Answer

A set of only noisy examples

Answer

The final exam never used until the end

Question 16

What is "early stopping" and how does it fight overfitting?

Accepted Answer

Stopping training when validation performance stops improving

Answer

Stopping training in the very first second

Answer

Never stopping training at all

Answer

Stopping only when training accuracy is low

Question 17

If a model has way more adjustable settings (parameters) than training examples, what risk grows?

Accepted Answer

Overfitting, because the model can memorize each example exactly

Answer

Underfitting, because it is too simple

Answer

No risk, more settings is always good

Answer

The model becomes faster

Question 18

How can showing a model more varied examples (data augmentation) reduce overfitting in image tasks?

Accepted Answer

It adds varied versions of data so the model learns features, not exact images

Answer

It deletes hard examples

Answer

It makes all images identical

Answer

It removes the need for testing

Question 19

Why might a model that scores 100% on training data actually worry an ML engineer?

Accepted Answer

A perfect training score may signal memorization and poor real-world performance

Answer

It always means the model is perfect

Answer

It means there is too little data

Answer

Perfect training scores are impossible

Question 20

What is the "signal" in data, and why do we want the model to learn it instead of the noise?

Accepted Answer

The true useful pattern; learning it helps the model handle new data

Answer

Random errors we want to copy

Answer

The slowest part of training

Answer

The name of a Python library

Question 21

How does "cross-validation" give a more trustworthy check against overfitting?

Accepted Answer

It tests on multiple different splits and averages for a reliable estimate

Answer

It uses only one fixed split forever

Answer

It trains without ever testing

Answer

It removes all the data

Question 22

A weather app trained only on summer data predicts badly in winter. Is this overfitting, and why?

Accepted Answer

Yes, it learned only summer patterns and cannot generalize to winter

Answer

No, it learned everything perfectly

Answer

No, weather cannot be predicted at all

Answer

Yes, but only because the app is slow

Question 23

What does "dropout" do in a neural network to help with overfitting?

Accepted Answer

It randomly disables some units during training to encourage robust learning

Answer

It deletes data permanently

Answer

It speeds up the internet

Answer

It copies the training answers

Question 24

Why is it unfair to judge a model's real skill using the same data it trained on?

Accepted Answer

The model already saw that data, so high scores may just be recall, not skill

Answer

Training data is always wrong

Answer

It is faster to reuse training data

Answer

Using training data makes the model smarter

Question 25

In simple terms, what is the "sweet spot" between underfitting and overfitting?

Accepted Answer

Complex enough to learn the pattern but simple enough to ignore noise

Answer

As complex as possible

Answer

As simple as possible

Answer

Whatever trains the fastest

Question 26

How can removing some unnecessary features (inputs) help reduce overfitting?

Accepted Answer

Fewer irrelevant features means fewer chances to fit noise

Answer

More features always reduce overfitting

Answer

Features have no effect on overfitting

Answer

Removing features always causes underfitting

Question 27

If your model does poorly on both training and test data, is that overfitting or underfitting?

Accepted Answer

Underfitting, because it fails even on the training data

Answer

Overfitting, because both scores are low

Answer

Perfect fitting

Answer

Neither, the data is broken

Question 28

Why does overfitting matter for fairness when AI is used to make real decisions?

Accepted Answer

An overfit model may work only for cases it saw, causing unfair results for others

Answer

Overfitting always makes AI perfectly fair

Answer

Fairness has nothing to do with data

Answer

Overfitting only affects training speed

Question 29

What simple comparison of two numbers best reveals overfitting?

Accepted Answer

Training accuracy versus test accuracy

Answer

Number of features versus number of rows

Answer

Training time versus file size

Answer

Model name versus dataset name

Question 30

How is overfitting like a student who only studies the exact practice questions and ignores the concepts?

Accepted Answer

Both ace familiar items but fail on slightly new ones due to memorizing

Answer

Both always perform perfectly everywhere

Answer

The student learns faster than the model

Answer

Neither has any problem

Test yourself