You are on page 1of 14

CLASSIFYING PERSONAL LOAN BY

SUPPORT VECTOR MACHINE

LA THANH THAO

Thesis Supervisor: Ph.D. PHAM HAI HA

April 26, 2019


TABLE OF CONTENT

I Problem
I Distance from a point to a hyperplan
I An optimal problem for SVM
Problem

Mr. John wants to get a loan from a local Bank. He is considered


to be accpedted or not based on his Age and Revenue.

What is the best Classification for a bank to avoid the default risk?
Problem

Figure: Separated training data set


Problem

Figure: The linearly saparable lines


Problem

Figure: A perfect Linearly Separable Figure: A bad Linearly Separable


Distance from a point to a hyperplane

Figure: Margin of solution

|w1 x1 + w2 x2 + b|
d(point, plane)= q
w12 + w22
Distance from a point to a hyperplane

.
|w1 x1 + w2 x2 + b|
I d(point, plane)= q
w12 + w22
|w1 x1 + w2 x2 + w3 x3 + b|
I d(point, plane)= q
w12 + w22 + w32
|w1 x0 + w2 x2 + ... + wn xn + b|
I d(point, plane)= q
w12 + w22 + ... + wn2
An optimal problem for SVM

I (x11 , x21 ), (x12 , x22 ), ..., (x1n , x2n )


with xi ∈ Rd
I d is the number of
dimention.
I N is the number of Data
I Signal of the trainning data
and yi (yi = 1 (class 1)
yi = −1 (Class 2))is the
label of that data.

Figure: Training data set


An optimal problem for SVM

|w1 x1 + w2 x2 + b|
I q
w12 + w22
|w1 x1 + w2 x2 + b|
Margin= min q
w12 + w22

Figure: Training data set


An optimal problem for SVM

I Distance from a closet point


|w1 x1 + w2 x2 + b|
Margin= min q
w12 + w22

I Maximize the margin


yi (w1 x1 + w2 x2 + b)
(w , b)= argmax min q
w12 + w22

1
argmax q minyi (w1 x1 + w2 x2 + b)
w12 + w22
An optimal problem for SVM

I yi (w1 x1 + w2 x2 + b) = 1

I yi (w1 x1 + w2 x2 + b) ≥ 1
when yi = 1
yi (w1 x1 + w2 x2 + b) ≤ −1
when yi = −1

yi (w1 x1 + w2 x2 + b) ≥ 1

Figure:
An optimal problem for SVM

1
I argmax q minyi (w1 x1 + w2 x2 + b)
w12 + w22

1
I (w,b)=argmax q
w12 + w22
Subject to: yi (w1 x1 + w2 x2 + b) ≥ 1

1
I (w,b)=argmin (w12 + w22 )
2
Subject to: yi (w1 x1 + w2 x2 + b) − 1 ≥ 0
Dual Quadratic programming

1
I (w,b)=argmin (w12 + w22 )
2
Subject to: yi (w1 x1 + w2 x2 + b) − 1 ≥ 0

I Larangian function
1
L(w1 , w2 , b, λ)= (w12 + w22 ) - N n n
n=1 λn (yi (w1 x1 + w2 x2 + b) − 1)
P
2

You might also like