SVM

Question 1

Idea behind SVM?
Best Hyper-Plane

Answer 1

To maximize the gap between plane and point so that classification happens as widely as possible.

Question 2

Why SVM?

Answer 2

• Linear & Non-Linear
• Classification, Regression & Outlier Detection (One Class SVM)

Question 3

What is One Class SVM?

Answer 3

Anomaly Detection technique

Separates majority class with minority using kernel trick.

Question 4

When SVM? (Real world examples)

Answer 4

Text classification, Image classification, Spam Detection, Handwriting identification, Gene expression analysis, Face detection and Anomaly detection

Question 5

What is Hyperplane?

Answer 5

In a p-dimensional space, a hyperplane is a sub-space of p-1 dimensions.

It is the plane that separates the support vectors.

Question 6

What is Support Vector?

Answer 6

Points touching the decision boundary (classifier).

Question 7

Maximum Margin Hyperplane/Hard Margin Hyperplane

Answer 7

Hyperplane whose distance from points on either side is maximum.

Question 8

Advantage of SVM

Answer 8

1. Robust to Outliers.
2. Can also be used for Non Linear Classification
3. Can also be used for Regression

Question 9

Why is the coefficient of the equation of the line of hyperplane on either side is 1?

Answer 9

Just to simplify the calculation because even if we take the values as K, we can divide the whole equation by K and get the eqn. with coefficient 1.

Question 10

Soft Margin classifier / Support vector classifier

Answer 10

When the data is not perfectly separable, misclassifications are allowed and a penalty term is added allowing for a trade-off between a wider margin and few misclassifications.

Question 11

SVM Error

Answer 11

Margin Error + Classification Error
i.e., Hinge Loss + Regularization Term (Penalty)

Question 12

What is “C” in Hinge Loss?

Answer 12

C is a hyper parameter which strikes a balance between making the gap as big as possible and reducing mistakes for misclassification.

Question 13

What is Kernel in SVM?

Answer 13

Kernel is a mathematical function which is used to map lower dimensional feature space into higher dimensional feature space.

Examples:- Linear, Polynomial, Radial Basis Function (RBF) and sigmoid.

Question 14

What happens with increase or decrease in “C” in SVM?

Answer 14

With increasing in C, the margin widens and with decrease in C, it become less tolerant of violations, and so the margin narrows.

C controls the bias-variance trade-off.

Question 15

What is the advantage of using a kernel rather than simply enlarging the feature space using functions of the original features?

Answer 15

Computational - kernel do not forms extra column for enlarged features, rather it transforms the existing features.

Question 16

SVM for more than two classes

Answer 16

1. One vs one - Kc2 SVM - voting count
2. One vs All - k SVM

Question 17

Is the concept of kernel only applicable to SVM?

Answer 17

No, We can apply nonlinear kernels in other classification technique as well. However, for historical reason, the use of kernel is widespread in context of SVM