Support vector machines quiz

Question 1

What is the main goal of a Support Vector Machine (SVM) classifier?

Accepted Answer

Find the hyperplane that maximizes the margin between classes

Answer

Minimize the number of features used

Answer

Cluster data into groups without labels

Answer

Fit a curve through every training point

Question 2

What are 'support vectors' in an SVM?

Accepted Answer

The data points closest to the decision boundary that define it

Answer

The points farthest from the boundary

Answer

All the points in the majority class

Answer

The mean vectors of each class

Question 3

What is the 'margin' in an SVM?

Accepted Answer

The distance between the hyperplane and the nearest data points

Answer

The total spread of the dataset

Answer

The error rate of the classifier

Answer

The number of misclassified points

Question 4

What is a hyperplane in the context of SVM?

Accepted Answer

A decision boundary one dimension lower than the feature space

Answer

A point at the center of the data

Answer

A curved surface that wraps around the data

Answer

A list of all support vectors

Question 5

What does the equation of an SVM decision boundary, w·x + b = 0, represent?

Accepted Answer

The separating hyperplane, classified by the sign of w·x + b

Answer

The margin width directly

Answer

The loss function of the model

Answer

The probability of each class

Question 6

What is the difference between a 'hard margin' and a 'soft margin' SVM?

Accepted Answer

Soft margin allows some violations; hard margin allows none

Answer

Hard margin uses kernels; soft margin does not

Answer

Soft margin only works in 3D

Answer

They are identical in practice

Question 7

What role does the regularization parameter C play in a soft-margin SVM?

Accepted Answer

It balances margin width against training errors

Answer

It sets the number of support vectors directly

Answer

It chooses which kernel to use

Answer

It scales the input features

Question 8

If you increase the value of C in an SVM, what generally happens to the margin?

Accepted Answer

The margin becomes narrower and overfitting risk rises

Answer

The margin becomes wider and underfitting risk rises

Answer

The margin is unaffected

Answer

The kernel automatically changes

Question 9

What is the 'kernel trick' in SVM?

Accepted Answer

Computing inner products in a high-dimensional space without explicit transformation

Answer

Deleting features to speed up training

Answer

Averaging multiple SVMs together

Answer

Reducing the dataset to two dimensions

Question 10

Why might you use a non-linear kernel instead of a linear SVM?

Accepted Answer

When classes are not linearly separable in the original space

Answer

When the dataset is very small

Answer

When all features are already binary

Answer

When you want fewer support vectors only

Question 11

What is the RBF (Radial Basis Function / Gaussian) kernel known for?

Accepted Answer

Modeling flexible non-linear boundaries based on point distances

Answer

Always producing linear boundaries

Answer

Requiring no parameters at all

Answer

Working only with text data

Question 12

In an RBF-kernel SVM, what does a large value of the gamma (γ) parameter do?

Accepted Answer

Each point has localized influence, giving wiggly boundaries and overfitting risk

Answer

It widens the margin automatically

Answer

It converts the kernel to linear

Answer

It reduces the number of features

Question 13

How does an SVM make a prediction for a new data point with a linear kernel?

Accepted Answer

By taking the sign of w·x + b

Answer

By counting the nearest neighbors

Answer

By computing the average of all labels

Answer

By measuring the dataset variance

Question 14

Is SVM primarily a supervised or unsupervised learning algorithm?

Accepted Answer

Supervised, requiring labeled data

Answer

Unsupervised, requiring no labels

Answer

Reinforcement learning

Answer

Semi-supervised only

Question 15

Why is feature scaling (standardization) important before training an SVM?

Accepted Answer

Because SVMs use distances, so unscaled features can dominate unfairly

Answer

Because SVMs cannot handle decimal numbers

Answer

Because scaling reduces the number of classes

Answer

Because it removes the need for a kernel

Question 16

Which scikit-learn class is commonly used to build an SVM classifier, and what does this snippet do?

Accepted Answer

SVC from sklearn.svm, which trains and predicts with an SVM

Answer

KMeans, which clusters the data

Answer

LinearRegression, which fits a line

Answer

PCA, which reduces dimensions

Question 17

What does the 'slack variable' represent in a soft-margin SVM?

Accepted Answer

The amount by which a point violates the margin

Answer

The kernel function value

Answer

The total number of features

Answer

The learning rate

Question 18

Can SVMs be used for regression tasks?

Accepted Answer

Yes, via Support Vector Regression (SVR)

Answer

No, SVMs only do classification

Answer

Only for binary outputs

Answer

Only with linear kernels

Question 19

How does an SVM naturally handle multi-class classification, given it is fundamentally binary?

Accepted Answer

By combining binary SVMs via one-vs-rest or one-vs-one

Answer

By using a single hyperplane for all classes

Answer

By clustering the classes first

Answer

SVMs cannot do multi-class at all

Question 20

What is a key advantage of SVMs in high-dimensional spaces (many features)?

Accepted Answer

They remain effective even when features outnumber samples

Answer

They require the number of features to be small

Answer

They ignore high-dimensional data entirely

Answer

They need at least 1000 samples per feature

Question 21

What is a notable disadvantage of SVMs on very large datasets?

Accepted Answer

Training time scales poorly with the number of samples

Answer

They cannot handle more than two features

Answer

They always overfit small datasets

Answer

They require no tuning at all

Question 22

What does it mean if a dataset is 'linearly separable'?

Accepted Answer

A straight line or flat hyperplane can perfectly separate the classes

Answer

The data forms a circle

Answer

No boundary can separate the classes

Answer

The data has only one class

Question 23

Why does the SVM optimization maximize the margin rather than just finding any separating boundary?

Accepted Answer

A larger margin generalizes better and is more robust to noise

Answer

A larger margin always reduces training time

Answer

Any boundary generalizes equally well

Answer

Smaller margins are always preferred

Question 24

In SVM, maximizing the margin is mathematically equivalent to minimizing what quantity?

Accepted Answer

Minimizing ½‖w‖² (the squared norm of the weight vector)

Answer

Maximizing the number of support vectors

Answer

Minimizing the bias term b

Answer

Maximizing ‖w‖

Question 25

What is the polynomial kernel used for in SVM?

Accepted Answer

Fitting curved boundaries via feature combinations up to a degree

Answer

Forcing the boundary to be a straight line

Answer

Removing non-linear relationships

Answer

Clustering the data into polynomials

Question 26

What is a valid kernel function required to satisfy (intuitively)?

Accepted Answer

It must correspond to an inner product in some feature space (Mercer's condition)

Answer

It must always return values above 1

Answer

It must be a linear function

Answer

It must ignore one of the inputs

Question 27

If two SVM models give the same training accuracy but one has a wider margin, which is generally preferred and why?

Accepted Answer

The wider-margin model, for better generalization

Answer

The narrower-margin model, for higher accuracy

Answer

Either one, margin does not matter

Answer

The one with more support vectors, always

Question 28

What does the 'decision function' value (distance from the hyperplane) tell you about an SVM prediction?

Accepted Answer

Its magnitude reflects how confident the prediction is

Answer

It gives the exact probability of error

Answer

It counts the support vectors

Answer

It equals the training accuracy

Question 29

How do you typically choose good values for C and gamma in an RBF SVM?

Accepted Answer

Via grid/random search with cross-validation

Answer

By always using the default values

Answer

By picking the largest possible values

Answer

By choosing values randomly without validation

Question 30

What happens to an SVM's decision boundary if you remove a non-support-vector training point?

Accepted Answer

The boundary stays unchanged

Answer

The boundary flips to the other side

Answer

The model must be retrained from scratch with a new kernel

Answer

The margin doubles

Test yourself