Cse352 AI

Homework 3
SOLUTIONS

PROBLEM 1
Use Lecture Notes to WRITE short, 1-2 paragraphs long ANSWERS
to the following questions.
Remember: on TEST your answers must match Lectures not
RANDOM Google pieces as a lot of students copied in. This is zero
points.

I compiled some answers by students that ARE acceptable.

1. Describe what is CLASSIFICATION; type of data, goals and

types of applications
Classification is an example of supervised learning in which
rules (characteristic and discriminant), or other patters
(converged Neural Network) are created based on training
data and tested with test data. If the predictive accuracy of
the rules (patters) is high, they can be used to classify new
data. The training data must be in the form of a table with its
key attribute removed and its class attribute distinguished.
The values in this class label must be discrete and
unordered. An example of an application of classification
could be customer. Profiling, where customer preferences
and which customers buy which products can be used to
identify the best products for different customers

2. Describe all stages of the classification process

The first step of the classification process as in any the
learning process is the selection of data. The selected data is
a subset of all the available data and must address the
problem at hand.
 The next step is to clean the selected data so that the impact
of missing, incomplete, or noisy data is minimized.
Next the data must continue to get preprocessed so that
learning algorithms can be applied to it.
Now a classification algorithm can be applied to the
preprocessed data to obtain patterns about the data.
In the final steps, the patterns are tested for their accuracy and
interpreted so that the user can decide if they still need refining
or can be presented.

3. Describe and discuss basic classification Models and their

differences

There are descriptive and statistical classification models.

Descriptive models are models are Decision Trees,
Classification by Association, Genetic Algorithms, and
Rough Sets. Statistical models are: Neural Networks,
Bayesian Networks, Support Vectors Model

4. Discuss the Decision Tree Induction and its strengths and

weaknesses

Decision  tree  is  constructed  in  a  top-­‐down  recursive  divide-­‐

and-­‐conquer  manner.  At  start,  all  the  training  examples  are  
at  the  root.  Attributes  are  categorical.  Examples  are  
partitioned  reclusively  based  on  selected  attributes.  Test  
attributes  are  selected  on  the  basis  of  a  heuristic  or  
statistical  measure.  Strength  of  Decision  Tree  method  is  
decision  trees  are  able  to  generate  understandable  rules  
and  performs  classification  without  requiring  much  
computation.  However,    trees  are  less  appropriate  for  
 estimation  tasks  where  the  goal  is  to  predict  the  value  of  a  
continuous  attribute  

5. Discuss the Neural Network Model and its strengths and

weaknesses

Neural network is a set of connected input/output units where each

connection has a weight associated with it. During the learning
phase, the network learns by adjusting the weights so as to be able
to predict the correct class label of the input tuples. Its weakness is
the long training time and it is difficult to interpret the symbolic
meaning behind the learned weights and of hidden units in the
network. Its strength is high tolerance to noisy data and ability to
classify incomplete data

6. Describe a process of building a CLASSIFIER

Building a classifier follows stages if any learning process.

The main two phases of building a classifier are: training
and testing. In both phase we use training data set and test
data. We use training data set to create patterns. We
evaluate created patterns with the use of test data. We
terminate the process if it has been trained and tested and
the predictive accuracy is on an acceptable level.

All stages are, like in any Learning Process the following

1) The first step is the selection of data. The selected data is a subset
of all the available data and must address the problem at hand.
2) The next step is to clean the selected data so that the impact of
missing, incomplete, or noisy data is minimized.
3) Next the data must continue to get preprocessed so that learning
algorithms can be applied to it.
4) Now a learning algorithm can be applied to the preprocessed data to
obtain patterns about the data.
 5) In the final steps, the patterns are tested for their accuracy and
interpreted so that the user can decide if they still need refining or
can be presented.

7. Define a CLASSIFIER

A classifier is the final product of a learning process that

uses a classification data set and a classification algorithm.
A CLASSIFIER is a black box that is used to classify
records for which the class label in unknown.

A classifier is the end result of a process that uses training data

and a classification algorithm to generate patterns that can
classify new data. The effectiveness, need for refinement, and
completion of a classifier is evaluated based on its predictive
accuracy. The patterns a classifier uses take different forms,
such as rules or trees, or trained different networks, like Neural
or Bayesian.

PROBLEM 2: BUILDING A CLASSIFIER

For the data set given below build a classifier following all steps
needed in the constructions:
preprocessing, training and testing
Describe and motivate your choice of algorithms and methods used at
each step.
CLASSIFICATION DATA:
Age Income Student Credit Rating Buys Computer

21 60,000 yes 3 No
30 70,000 No 5 No
38 38666.667 No 2 Yes
45 45,000 yes 3 Yes
46 25,000 no 2 Yes
47 30,000 Yes 6 No
39 28,000 Yes 5 No
29 48,000 Yes 3 No
50 75,000 Yes 2 No
48 55125 Yes 3 No
30 38666.667 Yes 6 Yes
51 46,000 No 4 Yes
32 80,000 Yes 2 No
45 50,000 No 4 No

PART 1: Preprocessing
Attributes: Age, Income, Credit Rating

1. Fill the missing Values

2. Use Binning Method to discretize values of attributes Age,

Income, Credit_Rating
The number of bins is up to you

1. Preprocessing Calculations
THIS is a Solution Submitted by a Student
There are many other solutions!
1. Mising Values: explain the method you used

Attribute values mean for income:

No Class
(60000+70000+30000+28000+48000+75000+80000+50000)/8=
441000/8=55125 Mean

Yes Class
(45000+25000+46000)/3=
116000/3=38666.667 Mean

2. Binninig: AGE

Equal-Depth binning for Age

N intervals
N=3
Depth=5

Bin 1 21 29 30 30 32
Bin 2 38 39 45 45 46
Bin 3 47 48 50 51

Bin 1: Age Group >=21&<=32

Bin 2: Age Group >32&<=46
Bin 3: Age Group >46&<=51

3. Binning: INCOME

Equal-Depth binning for Income

N intervals
N=3
Depth=5

Bin 1 25000 28000 30000 38666.667 38666.667

Bin 2 45000 46000 48000 50000 55125
Bin 3 60000 70000 75000 80000

Bin 1: >=25000&<=38666.667
Bin 2: >38666.667&<=55125
Bin 3: >55125&<=80000
4. Binning : CREDIT RATING

Equal-Width binning for Credit Rating

N intervals
N=2
Highest Value=6
Lowest Value=2
(6+2)/3=8/2=4

Bin 1 0>=&<4 2,2,2,2,3,3,3,3

Bin 2 4>=&<=8 4,4,5,5,6,6

Bin 1: >=0&<4
Bin 2: >=4&<=8

2. YOUR Data after Preprocessing:

Remark: Bins for different attributes should have and usually
HAVE special names to make the FINAL Descriptive Rules
comprehensible

Age Income Student Credit Rating Buys

Computer

Bin 1 Bin 3 yes Bin 1 No

Bin 1 Bin 3 No Bin 2 No
Bin 2 Bin 1 No Bin 1 Yes
Bin 2 Bin 2 yes Bin 1 Yes
Bin 2 Bin 1 no Bin 1 Yes
Bin 3 Bin 1 Yes Bin 2 No
Bin 2 Bin 1 Yes Bin 2 No
Bin 1 Bin 2 Yes Bin 1 No
Bin 3 Bin 3 Yes Bin 1 No
Bin 3 Bin 2 Yes Bin 1 No
Bin 1 Bin 1 Yes Bin 2 Yes
Bin 3 Bin 2 No Bin 2 Yes
Bin 1 Bin 3 Yes Bin 1 No
Bin 2 Bin 2 No Bin 2 No
3. Training and Testing
Specify which Decision Tree Algorithm you use
USE Three-Fold Cross Validation –Lecture 11testing

Remember: must perform training and testing 3 times and

predictive accuracy is averaged;
We adopt the union of rules as the
FINAL set of rules- YOUR Classifier

TRAINING- TESTING
ID3 non heuristic attributes chosen

REMARK THE TREE SHOULD BR CONSTRUCTED for

PREPROCESSED DATA- NOT ORIGINAL!!
Sets 1,2, 3 are original DATA NOT preprocessed DATA!
When you write SOLUTIONS by HAND, as on the TEST; you
MUST use preprocessed DATA!

When you use SOFTWARE, you input ORIGINAL DATA,

because SOFTWARE changes it accordingly to Preprocessing
program; so
THE TREE IS constructed (the student visibly used some
program!) for Preprocessed data, where BINS are replaced by
appropriate intervals. YOU CAN use software but MUST
present your data – and results correctly.

ON THE TEST you will have to do it all BY HAND.

Set 1:

Age Income Student Credit Rating Buys

Computer

21 60,000 yes 3 No
30 70,000 No 5 No
38 38,666.667 No 2 Yes
45 45,000 yes 3 Yes
46 25,000 No 2 Yes

Set 2:

Age Income Student Credit Rating Buys

Computer

47 30,000 Yes 6 No
39 28,000 Yes 5 No
29 48,000 Yes 3 No
51 46,000 No 4 Yes
32 80,000 Yes 2 No

Set 3:

Age Income Student Credit Rating Buys

Computer

50 75,000 Yes 2 No
48 55125 Yes 3 No
30 38666.667 Yes 6 Yes
45 50,000 No 4 No

Training 1

Age Income Student Credit Rating Buys Computer

 21 60,000 Yes 3 No
30 70,000 No 5 No
38 38,666.667 No 2 Yes
45 45,000 Yes 3 Yes
46 25,000 No 2 Yes
47 30,000 Yes 6 No
39 28,000 Yes 5 No
29 48,000 Yes 3 No
51 46,000 No 4 Yes
32 80,000 Yes 2 No

Age

>=21&<=32 >32&<=46 >46&<=51

Income Student Credit Buys Income Student Credit Buys

Rating Computer Rating Computer
60,000 Yes 3 No 38,666.667 No 2 Yes
70,000 No 5 No 45,000 yes 3 Yes
48,000 Yes 3 No 25,000 No 2 Yes
80,000 Yes 2 No 28,000 Yes 5 No

Income Student Credit Buys

Rating Computer
30,000 Yes 6 No
46,000 No 4 Yes

Age

>=21&<=32 >32&<=46 >46&<=51

Class = No
Credit Rating

>=0&<4 >=4&<=8

Income Student Buys

Computer
38,666.667 No Yes
45,000 yes Yes
25,000 No Yes Income Student Buys
Computer Student
28,000 Yes No
No Yes

Income Credit Buys

Rating Computer
46,000 4 Yes

Income Credit Buys

Rating Computer
30,000 6 No

Age

>=21&<=32 >32&<=46 >46&<=51

Class = No
Credit Rating

>=0&<4 >=4&<=8

Class = Yes
Class = No
No Student

Class = Yes Yes

Class = No

Rules : BINS names are replaced by bins intervals

R1. Age(x, >=21&<=32) ⇒ Class = buysComputer(x, No)
R2. Age(x, >32&<=46) Λ Credit Rating(x, >=0&<4) ⇒ Class = buysComputer(x,
Yes)
R3. Age(x, >32&<=46) Λ Credit Rating(x, >=4&<=8) ⇒ Class = buysComputer(x,
No)
R4. Age(x, >46&<=51) Λ Student(x, No) ⇒ Class = buysComputer(x, Yes)
R5. Age(x, >46&<=51) Λ Student(x, Yes) ⇒ Class = buysComputer(x, No)

Testing Set 3

Age Income Student Credit Rating Buys

Computer

50 75,000 Yes 2 No
48 55125 Yes 3 No
30 38666.667 Yes 6 Yes
45 50,000 No 4 No

Record 1 is well classified as of Rule 5

Record 2 is well classified as of Rule 5
Record 3 is misclassified
Record 4 is well classified as of Rule 3

3 records well classified out of 4

3/4*100%=75%= predictive accuracy

Training 2

Age Income Student Credit Rating Buys Computer

47 30,000 Yes 6 No
39 28,000 Yes 5 No
29 48,000 Yes 3 No
51 46,000 No 4 Yes
32 80,000 Yes 2 No
50 75,000 Yes 2 No
48 55125 Yes 3 No
30 38666.667 Yes 6 Yes
45 50,000 No 4 No
 Credit Rating

>=0&<4 >=4&<=8
Age Income Student Buys Age Income Student Buys
Computer Computer
29 48,000 Yes No 47 30,000 Yes No
32 80,000 Yes No 39 28,000 Yes No
50 75,000 Yes No 51 46,000 No Yes
48 55125 Yes No 30 38666.667 Yes Yes
45 50,000 No No

Credit Rating

>=0&<4 >=4&<=8
Class = No
Age
21>=&<=32 32>&<=46 46>&<=51
Age Student Buys Income Student Buys
Computer Income Student Buys
Computer
30 Yes Yes Computer
28,000 Yes No
30,000 Yes No
50,000 No No
46,000 No Yes
Credit Rating

>=0&<4 >=4&<=8
Class = No
Age
21>=&<=32 32>&<=46 46>&<=51
Class = Yes Class = No
Student
 No
Yes

Income Buys Income Buys

Computer Computer
46,000 Yes 30,000 No

Credit Rating

>=0&<4 >=4&<=8
Class = No
Age
21>=&<=32 32>&<=46 46>&<=51
Class = Yes Class = No
Student
No Yes
Class = Yes
Class = No

Rules

R1. Credit Rating(x, >=0&<4) ⇒ buysComputer(x, No)

R2. Credit Rating(x, >=4&<=8) Λ  Age(x, >=21&<=32) ⇒
buysComputer(x, Yes)
R3. Credit Rating(x, >=4&<=8) Λ  Age(x, >32&<=46) ⇒
buysComputer(x, No)
R4. Credit Rating(x, >=4&<=8) Λ  Age(x, >46&<=51) Λ Student(x, No)
⇒ buysComputer(x, Yes)
R5. Credit Rating(x, >=4&<=8) Λ  Age(x, >46&<=51) Λ Student(x, Yes)
⇒ buysComputer(x, No)
Testing Set 1

Age Income Student Credit Rating Buys

Computer

21 60,000 Yes 3 No
30 70,000 No 5 No
38 38,666.667 No 2 Yes
 45 45,000 Yes 3 Yes
46 25,000 No 2 Yes

Record 1 is well classified as of Rule 1

Record 2 is misclassified
Record 3 is misclassified
Record 4 is misclassified
Record 5 is misclassified
1 records well classified out of 5

1/5*100%=20%=predictive accuracy
Training 3

Age Income Student Credit Rating Buys Computer

21 60,000 Yes 3 No
30 70,000 No 5 No
38 38,666.667 No 2 Yes
45 45,000 Yes 3 Yes
46 25,000 No 2 Yes
50 75,000 Yes 2 No
48 55125 Yes 3 No
30 38666.667 Yes 6 Yes
45 50,000 No 4 No

Income
>=25000&<=38666.667 >55125&<=80000
>38666.667&<=55125

Age Student Credit Buys Age Student Credit Buys Age Student Credit Buys
Rating Computer Rating Computer Rating Computer
38 No 2 Yes 45 Yes 3 Yes 21 Yes 3 No
46 No 2 Yes 48 Yes 3 No 30 No 5 No
30 Yes 6 Yes 45 No 4 No 50 Yes 2 No

Income
>=25000&<=38666.667 >55125&<=80000
>38666.667&<=55125

Class = Yes Age Class = No

21>=&<=32 >32&<=46 46>&<=51

Student Credit Buys

Majority Rating Computer Student Credit Buys
Voting Yes 3 Yes Rating Computer
Class=No No 4 No Yes 3 No

Income
>=25000&<=38666.667 >55125&<=80000
>38666.667&<=55125

Class = Yes Age Class = No

21>=&<=32 >32&<=46 46>&<=51

Class = No Class = No
Student
No Yes
Credit Buys
Rating Computer
Credit Buys
4 No Rating Computer
3 Yes

Income

>=25000&<=38666.667 >55125&<=80000
>38666.667&<=55125

Class = Yes Age Class = No

21>=&<=32 >32&<=46 46>&<=51

Class = No Class = No
Student
No Yes

Class = No
 Class = Yes

R1. Income(x, >=25000&<=38666.667) ⇒ buysComputer(x, Yes)

R2. Income(x, >38666.667&<=55125) Λ  Age(x, >=21&<=32) ⇒ buysComputer(x,
No)
R3. Income(x, >38666.667&<=55125) Λ  Age(x, >32&<=46) Λ  Student(x,  No)  ⇒
buysComputer(x, No)
R4. Income(x, >38666.667&<=55125) Λ  Age(x, >32&<=46) Λ Student(x, Yes) ⇒
buysComputer(x, Yes)
R5. Income(x, >38666.667&<=55125) Λ  Age(x, >46&<=51) ⇒ buysComputer(x,
No)
R6. Income(x, >55125&<=80000) ⇒ buysComputer(x, No)

Training Set 2

Age Income Student Credit Rating Buys

Computer

47 30,000 Yes 6 No
39 28,000 Yes 5 No
29 48,000 Yes 3 No
51 46,000 No 4 Yes
32 80,000 Yes 2 No

Record 1 is misclassified
Record 2 is misclassified
Record 3 is misclassified
Record 4 is misclassified
Record 5 is misclassified

0 records well classified out of 5

0/5*100%=0%=predictive accuracy
4. MY CLASSIFIER

Union of Rules – must check for repetitions!

R1. Age(x, >=21&<=32) ⇒ Class = buysComputer(x, No)

R2. Age(x, >32&<=46) Λ Credit Rating(x, >=0&<4) ⇒ Class = buysComputer(x,

Yes)

R3. Age(x, >32&<=46) Λ Credit Rating(x, >=4&<=8) ⇒ Class = buysComputer(x,

No)

R4. Age(x, >46&<=51) Λ Student(x, No) ⇒ Class = buysComputer(x, Yes)

R5. Age(x, >46&<=51) Λ Student(x, Yes) ⇒ Class = buysComputer(x, No)

R6. Credit Rating(x, >=0&<4) ⇒ buysComputer(x, No)

R7. Credit Rating(x, >=4&<=8) Λ  Age(x, >=21&<=32) ⇒ buysComputer(x, Yes)

R8. Credit Rating(x, >=4&<=8) Λ  Age(x, >46&<=51) Λ Student(x, No) ⇒

buysComputer(x, Yes)

R9. Credit Rating(x, >=4&<=8) Λ  Age(x, >46&<=51) Λ Student(x, Yes) ⇒

buysComputer(x, No)

R10. Income(x, >=25000&<=38666.667) ⇒ buysComputer(x, Yes)

R11. Income(x, >38666.667&<=55125) Λ  Age(x, >=21&<=32) ⇒

buysComputer(x, No)
R12. Income(x, >38666.667&<=55125) Λ  Age(x, >32&<=46) Λ  Student(x,  No)  ⇒
buysComputer(x, No)

R13. Income(x, >38666.667&<=55125) Λ  Age(x, >32&<=46) Λ Student(x, Yes) ⇒

buysComputer(x, Yes)

R14. Income(x, >38666.667&<=55125) Λ  Age(x, >46&<=51) ⇒ buysComputer(x,

No)

R15. Income(x, >55125&<=80000) ⇒ buysComputer(x, No)

Predictive Accuracy = Training set one predictive accuracy+ Training set two
predictive accuracy+ Training set three predictive accuracy)/3

(75%+20%+0%)/3=95%/3=31.667% accuracy

REMARK: Predictive accuracy is very LOW

In Practice one would NOT accept this set of RULES as a CLASSIFIER!!

You must START the process over with different preprocessing!!

Problem 3: Learning Neural Networks

Given two records (Training Sample)

a1 a2 a3 Class
0.5 0 0.2 1
0 0.3 0 1

Use the Network below to evaluate a passage of TWO EPOCHS.

 1
w14
w15 4 w46
w24
2 6
w25
w56
w34 5
3 w35

Learning Rate: L= 0.7

REMEMBER: YOU HAVE TO SET YOUR INITIAL WEIGHTS AND BIASES

RANDOMLY; DON’T USE THE SET-UP FROM THE EXAMPLE.

FIRST EPOCH:

FirstrowofDATA:

a1 a2 a3 w14 w15 w24 w25 w34 w35 w46 w56 θ 4 θ5 θ6

0.5 0 0.2 0.1 0.2 0.3 0.4 0.1 0.1 0.5 0.2 0.3 0.2 0.1

Unitj Net Input Ij Output Oj

4 (0.5)(0.1)+(0)+(0.2)(0.1)+0.3= 0.37 1/(1+e-0.37) = 0.591

5 (0.5)(0.2)+(0)+(0.2)(0.1)+0.2= 0.32 1/(1+e-0.32) = 0.579

6 (0.5)(0.591)+(0.2)(0.579)+0.1= .511 1/(1+e-0.511) = 0.625

Error measurement:
Unitj Errorj
6 0.625(1-0.625)(1-0.625)=0.0878
We assume T6=1
5 0.579(1-0.579)(0.0878)(0.2)=0.0043

4 0.591(1-0.591)(0.0878)(0.5)=0.0153
New adjusted values:
Weight New Values
w46 0.5 + .7(0.0878)(.591) =0.536

w56 0.2 + .7(0.0878)(.579) =0.236

w34 0.1+ .7(0.0153)(.2) = 0.102

w35 0.1+ .7(0.0153)(.2) =0.102

w24 0.3+ .7(0.0153)(0) =0.3

 w25 0.4+ .7(0.0043)(0) =0.4

w14 0.1+ .7(0.0153)(0.5) =0.105

w15 0.2+ .7(0.0043)(0.5) =0.202

θ4 0.3 + .7(0.0153)=0.311

θ5 0.2+ .7(0.0043)=0.203

6 0.1+ .7(0.0878)=0.161

SecondRowofDATA:

a1 a2 a3 w14 w15 w24 w25 w34 w35 w46 w56 θ4 θ5 θ6

0 0.3 0 0.105 0.202 0.3 0.4 0.102 0.102 0.536 0.236 0.311 0.203 0.161

Unitj Net Input Ij Output Oj

4 (0)+(0.3)(0.3)+(0)+0.308= 0.398 1/(1+e-0.398) = 0.598

5 (0)+(0.3)(0.4)+(0)+0.203= 0.323 1/(1+e-0.323) = 0.58

6 (.536)(.598)+(.236)(.58)+0.164=0.62 1/(1+e-0.62) = 0.65

Error measurement:
Unitj Errorj
6 0.65(1-0.65)(1-0.65)=0.0796
We assume T6=1
5 0.58(1-0.58)(0.0796)(0.236)=0.0046

4 0.598(1-0.598)(0.0796)(0.536)=0.0103
 New adjusted values:
Weight New Values
w46 0.536 + .7(0.0796)(.598) =0.569

w56 0.236 + .7(0.0796)(.58) =0.268

w34 0.102 + .7(0.0103)(0) = 0.102

w35 0.102 + .7(0.0046)(0) =0.102

w24 0.3 + .7(0.0103)(0.3) =0.302

w25 0.4 + .7(0.0046)(0.3) =0.401

w14 0.105 + .7(0.0103)(0) =0.105

w15 0.202 + .7(0.0046)(0) =0.202

4 0.311 + .7(0.0103)=0.318

5 0.203 + .7(0.0046)=0.206

6 0.161 + .7(0.0796)=0.217

SECOND EPOCH:

FirstrowofDATA:

a1 a2 a3 w14 w15 w24 w25 w34 w35 w46 w56 θ4 θ5 θ6

0.5 0 0. 0.105 0.202 0.302 0.40 0.102 0.102 0.569 0.268 0.318 0.206 0.217
2 1

Unitj Net Input Ij Output Oj

4 (0.5)(0.105)+(0)+(0.2)(0.102)+0.318= 0.391 1/(1+e-0.391) = 0.597

5 (0.5)(0.202)+(0)+(0.2)(0.102)+0.206= 0.327 1/(1+e-0.327) = 0.581

6 (0.5)(0.597)+(0.2)(0.581)+0.217= 0.632 1/(1+e-0.632) = 0.653

Error measurement:
Unitj Errorj
6 0.653(1-0.653)(1-0.653)=0.079
We assume T6=1
5 0.581(1-0.581)(0.079)(0.268)=0.0052

4 0.597(1-0.597)(0.079)(0.569)=0.0108

New adjusted values:

Weight New Values
w46 0.569 + .7(0.079)(.597) =0.602

w56 0.268 + .7(0.079)(.581) =0.3

w34 0.102 + .7(0.0108)(.2) = 0.104

w35 0.102 + .7(0.0052)(.2) =0.103

w24 0.302 + .7(0.0108)(0) =0.302

w25 0.401 + .7(0.0052)(0) =0.401

w14 0.105 + .7(0.0108)(0.5) =0.109

w15 0.202 + .7(0.0052)(0.5) =0.204

θ4 0.318 + .7(0.0108)=0.326

θ5 0.206 + .7(0.0052)=0.210

θ6 0.217 + .7(0.079)=0.272

SecondRowofDATA:

a1 a2 a3 w14 w15 w24 w25 w34 w35 w46 w56 4 5 6

0 0.3 0 0.019 0.204 0.302 0.401 0.104 0.103 0.602 0.3 0.326 0.210 0.262
Unitj Net Input Ij Output Oj
4 (0)+(0.3)(0.302)+(0)+0.326= 0.417 1/(1+e-0.417) = 0.603

5 (0)+(0.3)(0.401)+(0)+0.210= 0.330 1/(1+e-0.330) = 0.582

6 (.602)(.603)+(.3)(.582)+0.262=0.8 1/(1+e-0.8) = 0.69

Error measurement:
Unitj Errorj
6 0.69(1-0.69)(1-0.69)=0.066
We assume T6=1
5 0.582(1-0.582)(0.066)(0.3)=0.0048

4 0.69(1-0.69)(0.066)(0.602)=0.0085

New adjusted values:

Weight New Values
w46 0.602 + .7(0.066)(.603) =0.569

w56 0.3 + .7(0.099)(.582) =0.268

w34 0.104 + .7(0.0085)(0) = 0.102

w35 0.103 + .7(0.0048)(0) =0.102

w24 0.302 + .7(0.0085)(0.3) =0.304

w25 0.401 + .7(0.0048)(0.3) =0.402

w14 0.019 + .7(0.0085)(0) =0.105

w15 0.204 + .7(0.0048)(0) =0.202

θ4 0.326 + .7(0.0085)=0.332

θ5 0.210 + .7(0.0048)=0.213

θ6 0.262 + .7(0.066)=0.308
PROBLEM 4: Classification by Association

Given a classification TRAINING data

Income Student Rating

high No Fair
low No Excellent
high yes Fair
medium No Fair
low Yes Fair
medium no Excellent

and classification TEST data

Income Student Rating

low Yes Excellent
medium No Fair
low no Fair
medium Yes Fair

Build a CLASSIFIER using Classification by Association Method

STEP 1: TRANSLATION – DATA TRANSFORMATION

Income  =   Income  =   Income  =   Student  =   Student  =   Rating  =   Rating  =  

Low  (I1)   medium   High  (I3)   No   (I4)   Yes   (I5)   Fair  (I6)   Excellent  
(I2)   (I7)  

-­‐   -­‐   +   +   -­‐   +   -­‐  

+   -­‐   -­‐   +   -­‐   -­‐   +  

-­‐   -­‐   +   -­‐   +   +   -­‐  

 -­‐   +   -­‐   +   -­‐   +   -­‐  

+   -­‐   -­‐   -­‐   +   +   -­‐  

-­‐   +   -­‐   +   -­‐   -­‐   +  

Let   minimum   support  count  be   2,  

Since  we   have  6  records  min  support  =  2/6   =  30%  

Let   minimum   confidence   required  be   70%  

Generating   1-­‐item  set   frequent  patterns  

Item  Set   Support  count  

I1   2  

I2   2  

I3   2  

I4   4  

I5   2  

I6   4  

I7   2  
C1   =  L1  

All   sc  >=   min  sc  

Item  Set   Support  count  

{I1,   I2}   0  

{I1,   I3}   0  

{I1,   I4}   1  

{I1,   I5}   1  
 {I1,   I6}   1  

{I1,   I7}   1  

{I2,   I3}   0  

{I2,   I4}   2  

{I2,   I5}   0  

{I2,   I6}   1  

{I2,   I7}   1  

{I3,   I4}   1  

{I3,   I5}   1  

{I3,   I6}   2  

{I3,   I7}   0  

{I4,   I5}   0  

{I4,   I6}   2  

{I4,   I7}   2  

{I5,   I6}   2  

{I5,   I7}   0  

{I6,   I7}   0  
C2  

Choose  candidates   with  sc  >=   min  sc  

Item  Set   Support  count  

{I2,   I4}   2  

{I3,   I6}   2  

{I4,   I6}   2  
 29

{I4,   I7}   2  

{I5,   I6}   2  
L2  

Joining  &  pruning  

Join   L2*L2  to  obtain  C3  
NO C3 Candidates

 
FOR  Classification  by  Association  we    only  
Consider  I={IX,IY}  Where  IY  represents      CLASS  
ATTRIBUTES:  
Rating  =  Fair   or  Rating  =  Excellent  

Item  Set   Support  count  

{I3,   I6}   2  

{I4,   I6}   2  

{I4,   I7}   2  

{I5,   I6}   2  
 30

Classification by Association Rules for TESTING

R1: IF Income(x, High) THEN Rating(x, Fair) [30, 70]

R2: IF Student(x, yes) THEN Rating(x, Fair) [30, 70]

TEST data

Income Student Rating

low Yes Excellent
medium No Fair
low no Fair
medium Yes Fair
 31

Record 1 –error, record 2- error, record 3- error,

record 4 – correctly classified by R2

Predictive accuracy 25% and 75% error rate

THE CLASSIFIER: - There is no Classifier

Predictive Accuracy with given test data
is too low to Accept the Rules R1, R2
as Classifier

START the Process OVER with different set of paramethers (or

DATA) if you have a choice