Scribd
Upload
44 views

Decision Trees Example

The document is an aptitude test determination sheet that evaluates 15 candidates across 5 tests: numeric, language literacy, deductive reasoning, abstract reasoning, and inductive reasoning. Each test section grades the candidates as pass, average, or fail. 9 of the 15 candidates were hired. The entropy of the overall system is calculated to be 0.971. The information gain of each test section is then calculated based on reducing the entropy. Based on the highest information gains, the optimal decision tree order is: abstract reasoning, numeric test, inductive reasoning, deductive reasoning, then language literacy.

Uploaded by

unwaz
Copyright
© © All Rights Reserved
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
44 views

Decision Trees Example

The document is an aptitude test determination sheet that evaluates 15 candidates across 5 tests: numeric, language literacy, deductive reasoning, abstract reasoning, and inductive reasoning. Each test section grades the candidates as pass, average, or fail. 9 of the 15 candidates were hired. The entropy of the overall system is calculated to be 0.971. The information gain of each test section is then calculated based on reducing the entropy. Based on the highest information gains, the optimal decision tree order is: abstract reasoning, numeric test, inductive reasoning, deductive reasoning, then language literacy.

Uploaded by

unwaz
Copyright
© © All Rights Reserved
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
You are on page 1/ 3

P52/8****/2016

Aptitude test determination sheet


Numeric
Test

Langua
ge
Literacy

P
F
A
F
P
P
F
P
A
P
F
A
A
F
A

P
F
A
P
A
P
F
P
P
F
A
A
F
F
A

Deducti
ve
Reasoni
ng
P
P
P
P
F
P
F
P
F
P
P
P
F
P
F

Abstract
Reasonin
g
P
P
P
P
P
P
F
F
P
P
P
F
P
P
F

Inductiv
e
Reasoni
ng
P
F
P
F
P
F
F
F
F
P
P
P
P
P
F

Hired

Departm
ent

Y
N
Y
N
Y
Y
N
Y
N
Y
Y
Y
N
Y
N

Audit
ICS
Audit
Tax
Advisory
Audit
ICS
Tax
ICS

KEY
P Pass, A Average, F Fail
NB: The numeric and language section of the test is graded as pass, average or fail
while the reasoning sections are graded in as pass or fail only.
Entropy (S) =

c Pi log2 Pi

i=1

Portion of the hired


9/15
Portion of the non-hired 6/15
Entropy (S) = (-9/15 log2 9/15) + (-6/15 log2 6/15)
= (-0.6 x -0.737) + (-0.4 x -1.322)
= 0.971
Gain (A,S) = Entropy(S) Numeric Test
NumericTest, Pass
NumericTest, Average
NumericTest, Fail

n |SAi|/|S| Entropy(SAi)

i=1

(Y=5, N=0)
(Y=2, N=3)
(Y=2, N=3)

Gain (NumericTest, S)
= 0.971 ( (5/15 ((0) + (0))) + (5/15((-2/5log22/5) + (-3/5log23/5))) + (5/15((2/5log22/5) + (-3/5log23/5))) )

= 0.971 - {[0] + [0.3333 (0.5288+0.4422)] + [0.3333 (0.5288+0.4422)]}


= 0.3238
Language Literacy
LanguageTest, Pass
LanguageTest, Average
LanguageTest, Fail

(Y=3, N=2)
(Y=4, N=1)
(Y=2, N=3)

Gain (LanguageTest, S)
= 0.971 ( (5/15 ((-3/5log23/5) + (-2/5log22/5))) + (5/15 ((-4/5log24/5) + (1/5log21/5))) + (5/15((-2/5log22/5) + (-3/5log23/5))) )
= 0.971 - {[0.3333(0.4422+0.5288)] + [0.3333(0.2575+0.4644)] +
[0.3333(0.5288+0.4422)]}
= 0.0832
Deductive Reasoning
DeductiveTest, Pass
DeductiveTest, Fail

(Y=8, N=2)
(Y=1, N=4)

Gain (DeductiveTest, S)
= 0.971 ( (10/15 ((-8/10log28/10) + (-2/10log22/10))) + (5/15 ((-1/5log21/5) + (4/5log24/5))) )
= 0.971 - {[0.6667(0.2575+0.4644)] + [0.3333(0.4644+0.2575)]}
=0.2491
Abstract Reasoning
AbstractTest, Pass
AbstractTest, Fail

(Y=5, N=0)
(Y=2, N=3)

Gain (AbstractTest, S)
= 0.971 ( (11/15 ((0) + (0))) + (4/15((-2/5log22/5) + (-3/5log23/5))) )
=0.971 - {[0] + [0.2667(0.5288+0.4422)]}
=0.712
Inductive Reasoning
InductiveTest, Pass
InductiveTest, Fail

(Y=7, N=1)
(Y=2, N=5)

Gain (InductiveTest, S)
= 0.971 ( (8/15 ((-7/8log27/8) + (-1/8log21/8))) + (7/15((-2/7log22/7) + (5/7log25/7))) )
= 0.971 - {[0.5333(0.1685+0.375)] + [0.4667(0.5164+0.3467)]}
= 0.2784

From the gains calculated, the decision tree will be of the form:
0.712 -> 0.3238 -> 0.2784 -> 0.2491 -> 0.0832
Abstract Reasoning
|
V
Numeric Test
|
V
Inductive Reasoning
|
V
Deductive Reasoning
|
V
Language Literacy

You might also like