Decision Tree

Choice of the Best Alternative

Manahan Siallagan
 Review

We discussed how to set “what is problem”:

How to set objectives;
How to set criteria and alternatives;
How to choose alternative without uncertainties
How to construct utility function
 Decision Analysis Problems
Introducing new product into marketplace:

What will the

reaction of
potential
customers ?
Competitors?

Tools:
Decision
Tree
3
 Decision Analysis Problems:
Investing in securities :
• How is the
economy?
• How about
interest
rates?

Tools:
Decision
Tree
4
 Decision Analysis Problems:
Drilling for oil in particular location :

• How likely is
there to be oil in
that location?
• How much?

Tools:
Decision
Tree
5
 Problems of Choice

C1 O1
D
C2 O2

w ay
be st  D : a decision maker
e
Th
 C : possible courses of action
(alternatives)
 O : desirable outcome;
1
6  O : undesirable outcome
2
 Complex Rational Choice

The information and insight

needed to predict the I can’t see
consequences of each option

Incomplete information
The other side may influences the
consequences (Strategic/Interactive
Decision).

It is assumed that we can calculate

probability of the consequences

7
Decision Analysis Problems
 Decision analysis is designed to address
decision making in the face of great
uncertainty
 Introducing new product into
marketplace:
 What will the reaction of potential
customers ? Competitors?
 Investing in securities :
 How is the economy?
 How about interest rates?
 Selecting the mix of crops and livestock
for the upcoming season :
 What will be the weather conditions?
 Drilling for oil in particular location :
 How likely is there to be oil in that
location?
 How much?
 Tools: Decision tree
 Why Decision Tree?
It can help a decision maker to develop a clear view of
the structure of a problem anda make it easier to
determine the possible scenarios that can result if a
particular course of action is chosen.
Decision trees can also help a decision maker to judge
the nature of the information that needs to be gathered
in order to tacke a problem.
It can be an excellent medium for communicating one
person’s perception of a problem to other individuals.

9
 Model of Decision Tree
p1,1
E1,1 O1
C1 p1,2
E1,2 O2
D E2,1 p2,1 O1
C2
E2,2 O2
p2,2
 D : a decision maker
 C : possible courses of action (alternatives)
 O1 and O2 : possible outcomes/consequences/payoffs
 Ei,j : Events (State of Natures/SON)
 pi,j : probabilities
10
 Structure of Decision Tree
Decision Node
Alternatives available for decision maker to choose;
Situation controllable by decision maker.

Alternatives of actions
11
 Structure of Decision Tree
Event Node
Events may happen after every action made by decision
maker;
Uncontrollable by decision maker;
Decision maker only has information about probability
of each event  no complete information.

p1
E1

E2 p2
An action
E3
p3

12 Events
 Building Decision Tree

1. Identify what decisions should be made by DM;

 What are the first decision, and next decisions to
be made?
2. Identify what SON happen after each decision;
3. Draw decision node and event node (SON);
4. Complete information about probabilities;
5. Complete information about payoff.

13
Goferbroke Company Case
 Goferbroke Company(1)
 Max Flyer is the founder of and sole owner of the Goferbroke Company,
which develops oil wells in unproven territory. Max’s friends refer to him
affectionately as a wildcatter. However he prefers to think himself as an
entrepreneur. He has poured his life saving’s into the company in the
hope of making it big with a large strike of oil.
 Now his chance possibly has come. His company has purchased various
tracts of land that larger oil companies have spurned as unpromising even
though they are near some large oil fields. Now Max has received an
exciting report about one of these tracts. A consulting geologist has just
informed Max that he believes there is one chance in four of oil there.
 Max has learned from bitter experience to be skeptical about the chances
of oil reported by consulting geologist. Drilling for oil on this tract would
require an investment of about $100,000. If the lands turns out to be dry
(no oil), the entire investment would be lost. Since his company doesn’t
have much capital left, this lost would be quite serious.
Goferbroke Company(2)
 On the other hand, if the tract does contain oil, the consulting geologist
estimates that there would be enough there to generate a net revenue of
approximately $800,000, leaving an approximate profit of:
 Profit if find oil = Revenue if find oil – Drilling cost
= $800,000 - $100,000
= $700,000

 There is another option that another oil company has gotten wind of
consulting geologist’s report and so has offered to purchase the tract of
land from Max for $90,000. This is very tempting. This too would
provide a welcome infusion of capital into the company, but without
incurring the large risk of a very substansial loss of $100,000.
 The Goferbroke Company Problem
An E P(E)=p  P(E’)= 1-p P(E)+P(E’)=1

 Decision that must be taken:

Should Max sell his land or doing drilling?
 Alternative:
1. Sell land
2. Drilling
 Possibility event that could happen (state of nature):
 Found Oil
 No Oil (dry)
 Payoff Table (Information from the case) Payoffs

If you are Max, which alternatives that you would

choose?
 Decision Criterion
If you are Max, which alternatives that you would
choose?
MAXIMAX = Maximum of maximum
Optimism  risk taker

700
700 Drill
90

MAXIMIN = Maximum of minimum

pessimism  risk avoidance

-100
90 Sell
90
 Decision Criterion
If you are Max, which alternatives that you would
choose?
Likelihood

Sell

High probability
 Decision Criterion
If you are Max, which alternatives that you would
choose?
Bayes EV (Expected Value)
EV(Alternative) = Sigma p(event)* payoff

EV(Drill) = Sigma p(event)* payoff

= (0.25)*(700) + (0.75)*(-100)
= 100

EV(Sell) = Sigma p(event)* payoff Max{EV(drill),EV(sell)}

= (0.25)*(90) + (0.75)*(90) Max{100,90}=100 Drill
= 90
 Classical probability:
Toss the coin  U={H,T}  n(U)=2
P(H) = n(H)/n(U) = ½
Set of card  U(C)={JH,JD,JS,JC,….} n(C)=52
P(ACE)=4/52
P(ACE and RED) = 2/52
Experimental Probability:
Toss 500 times coin  n(H)= 265, n(T)=235 n(E)=500
P(H) = 265/500; P(T)=235/500
Subjective Probability:
Expert

21
The Maximax Decision Criterion
 Focus only on the best that can happen “ the
maximax criterion always chooses the decision
alternative that can give the largest possible
payoff “Total Optimist
 Identify the maximum payoff from any SoN for
each decision alternative
 Find the maximum of these maximum payoffs
and choose the corresponding decision
alternative
 Weakness : abandoning prior probability and
abandoning other payoff beside only the biggest.

Maximax
The Maximin Decision Criterion
 Focus only on the worst that can happen to us
(total pessimist)
 Identify the minimum payoff from any SoN for each
decision alternatives
 Find the maximum of these minimum payoffs and
choose the corresponding decision alternative
 Weakness : abandoning prior probability and
abandoning other payoff beside only the maximin.

Maximin
The Maximum Likelyhood Criterion
 Focus on the most likely state of nature.
 Identify the state of nature with the largest
prior probability;
 Choose the decision alternative that has
the largest payoff for this state of nature.
 Weakness: abandoning payoff, that
actually payoff maybe very big.

Step 2 : Maximum

Step 1 : Maximum
Bayes’ Decision Rule
 Bayes’ Decision Rule choose best alternative by considering
entire information that being owned by doing steps mentioned:
 Calculate Expected Value for every decision alternative
EV = (prior prob x payoff)
 Choose the decision alternative that has the largest Expected
Value
 Advantage :
 Considering entire information (alternatives, payoffs, and prior
probabilities);
 In long term, if the decision occur sequential, then this criteria
will resulting payoff that mostly probably happen.

Maximum
Decision Trees of Goferbroke

Decision tree is decision making help

tools that could describe entire
alternatives with whole events that may
happen (SoN).
Showing : Alternatives, SoN, Prior
Probability, and Payoff.
Using Bayes’ Decision Rule to choose the
best action.
DT of Goferbroke’s Case
 Decision:
Drill or Sell the Land
 SON
Oil or Dry

Decision nodes Drill

-100

Sell

90
DT of Goferbroke’s Case
Event nodes

Oil 0.25
800
Drill

-100 Dry 0,75

0
Sell

90
DT of Goferbroke’s Case
Payoff

Oil 0.25
700
800
Drill

-100 Dry 0,75

-100
0
Sell
90
90
DT of Goferbroke’s Case
Payoff

Oil 0.25
700
800
Drill

-100 100 Dry 0,75

-100
0
Sell
100 Expected payoff
90 = MAX [100,90]
90 90
= 100
Expected Value (EV) per event node;

Action: Drill

100=(0.25*700) + (0.75*(-100))
 Decision Analysis:
New Information or Posterior
Probability

31
 Process in Revising Decision tree
Prior probability

New information

Posterior probability

32
Assessing Probability
 There are three approaches to assessing the probability of an
uncertain event:
1. a priori classical probability

2. empirical classical probability

3. subjective probability
an individual judgment or opinion about the probability of occurrence
 Computing Joint and
Marginal Probabilities

The probability of a joint event, A and B:

Computing a marginal (or simple) probability:

 Where B , B , …, B are k mutually exclusive and collectively

1 2 k
exhaustive events
 P(Ace) = 4/52
P(Ace) = P(Ace and Red) + (Ace and Black)
 = 2/52 + 2/52 = 4/52

35
Joint Probability Example

P(Red and Ace)

Color
Type Red Black Total
Ace 2 2 4
Non-Ace 24 24 48
Total 26 26 52
Marginal Probability Example

P(Ace)

Color
Type Red Black Total
Ace 2 2 4
Non-Ace 24 24 48
Total 26 26 52
 Computing Conditional Probabilities
A conditional probability is the probability of one event,
given that another event has occurred:
The conditional
probability of A given
that B has occurred

The conditional
probability of B given
that A has occurred

Where P(A and B) = joint probability of A and B

P(A) = marginal probability of A
P(B) = marginal probability of B
 Conditional Probability Example

 Of the cars on a used car lot, 70% have air

conditioning (AC) and 40% have a CD player (CD).
20% of the cars have both.
P(A and B) = P (B and A)
What is the probability that a car has a CD player, given that
it has AC ?

i.e., we want to find P(CD | AC) = P(CD and AC)/P(AC)

P(AC) =0.7
P(CD) 0.4
P(CD | AC) = P(CD and AC)/P(AC) P(CD and AC) =0.2
= 0.2/0.7
 Conditional Probability Example
(continued)
 Of the cars on a used car lot, 70% have air conditioning (AC) and 40% have a
CD player (CD).
20% of the cars have both.

CD No CD Total

AC 0.2 0.5 0.7

No AC 0.2 0.1 0.3
Total 0.4 0.6 1.0
 Conditional Probability Example
(continued)
 Given AC, we only consider the top row (70% of the cars). Of these, 20%
have a CD player. 20% of 70% is about 28.57%.

CD No CD Total

AC 0.2 0.5 0.7

No AC 0.2 0.1 0.3
Total 0.4 0.6 1.0
 Using Decision Trees
Given AC or no P(AC and CD) = 0.2
a s CD
AC: H
) = 0.7
(A C D o es
P have not P(AC and CD’) = 0.5
A C CD
Has
All
Cars
Doe
hav s not
eA P(AC’ and CD) = 0.2
C P(A a s CD
C’) H
= 0.3
D o es
have not P(AC’ and CD’) = 0.1
CD
 Using Decision Trees
(continued)

Given CD or no C P(CD and AC) = 0.2

Ha s A
CD:
0 .4
D ) = D o es
P (C have not P(CD and AC’) = 0.2
C D AC
Has
All
Cars
Doe
hav s not
eC C P(CD’ and AC) = 0.5
D P (C a s A
D’ ) H
=0
.6
D o es
have not P(CD’ and AC’) = 0.1
AC
 Bayes’ Theorem
Bayes’ theorem is used to revise previously calculated probabilities after new
information is obtained

 where:
Bi = ith event of k mutually exclusive and collectively
exhaustive events
A = new event that might impact P(Bi)
Goferbroke’s Case Continued
 Survey by geologist will provide more accurate
information about P(oil);
 How if Max has to decide two alternatives:
1. Do survey before drill/sell
2. Drill/sell without Survey
 Events:
 Do Survey
 FSS : Favorable Seismic Sounding : Oil is fairly likely
 USS : Unfavorable seismic sounding: Oil is quite unlikely.
 Drill or Sell
 Oil
 Dry
Max`s Experience
•P(state) prior; which is P(Oil)=0.25 & P(Dry)=0.75;
•P (finding|state) being known based on Max’s
experiences; which is
• P(FSS|Oil)=0.6,
• P(USS|Oil)=0.4,
• P(FSS|Dry)=0.2, and
• P(USS|Dry)=0.8

Which:
• State : Oil and Dry;
• Finding : FSS and USS;
• FSS : favorable seismic sounding; oil is fairly likely;
• USS : unfavorable seismic sounding; oil is quite unlikely.
 Max`s Experience
•P(state) prior; which is P(Oil)=0.25 & P(Dry)=0.75;
• P(A) + P(A’) = 1  P(A’) 1- P(A)
•P (finding|state) being known based on Max’s
experiences; which is
• P(FSS|Oil)=0.6,  P(USS|Oil)=0.4
• P(FSS|Dry)=0.2,  P(USS|Dry) =0.8

P(Oil|FSS) = P(Oil and FSS)/P(FSS)

P(FSS|Oil)= P(FSS and Oil)/ P(Oil) = 0.6 =0.15/0.3 = 0.5
P(FSS and Oil) = P(FSS|Oil) * P(Oil) = 0.6 * 0.25 = 0.15
P(FSS|Dry)= P(FSS and Dry)/P(Dry)=0.2
P(FSS and Dry) = P(FSS|Dry) * P(Dry) = 0.2 * 0.75 = 0.15

P(FSS)= P(FSS and Oil) + P(FSS and Dry) = 0.15 + 0.15 = 0.3
 Table:

FSS USS
Oil P(oil and FSS) P(Oil and USS) P(Oil)
Dry P(Dry and FSS) P(Dry and USS) P(Dry)
P(FSS) P(USS) 1

FSS USS
Oil 0.15 (0.25-0.15)=0.1 0.25
Dry 0.15 ().75-0.15)=0.6 0.75
0.3 0.7 1

48
 P=?
Oil

Drill

P=?
P=? Dry
Unfavorable

Sell

Do Survey

30000
P=? 0.5
Oil

Drill

P=?
P=? 0.3 Dry 0.5
Favorable
1

Sell

P=?
Oil

Drill

P=?
Dry
No Survey

Sell
Max`s Experience
•P(state) prior; which is P(Oil)=0.25 & P(Dry)=0.75;
•P (finding|state) being known based on Max’s
experiences; which is
• P(FSS|Oil)=0.6,
• P(USS|Oil)=0.4,
• P(FSS|Dry)=0.2, and
• P(USS|Dry)=0.8

Which:
• State : Oil and Dry;
• Finding : FSS and USS;
• FSS : favorable seismic sounding; oil is fairly likely;
• USS : unfavorable seismic sounding; oil is quite unlikely.
 P(Oil│USS)
Oil

P(FSS│Oil) = 0,6 Drill

P(USS│Oil) = 0,4 P(Dry│USS)

P(USS) Dry
P(FSS│Dry) = 0,2 Unfavorable
1
P(USS│Dry) = 0,8
Sell

Do Survey
P(Oil│FSS)
Oil

Drill

P(Dry│FSS)
P(FSS) Dry
Favorable

Sell

P(Oil)
Oil

Drill

P(Dry)
Dry
No Survey

Sell
Posterior Probability Formula

P(FSS|Oil) = P(FSS&Oil) / P(Oil)

P(FSS|Dry) = P(FSS&Dry) / P(Dry)
P(Oil|FSS) = P(Oil&FSS) / P(FSS)
P(Dry|FSS) = P(Dry&FSS) / P(FSS)
P(Oil|USS) = P(Oil&USS) / P(USS)
P(Dry|USS) = P(Dry&USS) / P(USS)
P(FSS&Oil) = P(Oil&FSS)  Law of Probability
Contingency table

FSS USS P(FSS|Oil) = P(FSS& Oil) / P(Oil)

Since :
P (Oil) = 0.25; P(FSS|Oil) = 0.6
Oil 0.15 0.1 0.25 Then
P(FSS & Oil) = P(FSS|Oil) x P(Oil)
Dry 0.15 0.6 0.75 = 0.6 X 0.25
= 0.15
0.3 0.7 Do same step for find P (FSS&Dry) = 0.15
So :
P(FSS) = P(FSS&Oil)+P(FSS&Dry)
= 0.3
Posterior Probability Formula (cont’d)
P(Oil|FSS) = P(Oil&FSS) / P(FSS)
Since : P(FSS) = 0.3;  from contigency table
P(Oil&FSS) = P(FSS&Oil) = 0.15
Then : P(Oil|FSS) = 0.15 / 0.3 = 0.5
Do same step for P(Oil|FSS); P(Dry|FSS); P(Oil|USS);
P(Dry|USS)
 FSS USS P(Oil|FSS)= P(Oil&FSS) = 0.15 = 0.5
P(FSS) 0.3
Oil 0.15 0.1 0.25

Dry 0.15 0.6 0.75 P(Dry|FSS)= P(Dry&FSS) = 0.15 = 0.5

P(FSS) 0.3
0.3 0.7

Do same step for P(Oil|USS); P(Dry|USS)

55
 Leveled Decision Analysis
P(Oil|USS)
P(USS)
P(Dry|USS)

Expected payoff
= MAX [123,100]
= 123 P(Oil|FSS)
P(FSS)
P(Dry|FSS)

P(Oil)

P(Dry)
Posterior Probability
 Given:
 P(state)=prior probability: P(oil) and P(dry)
 P(finding|state) = Max’s experience on probabilities of finding (FSS or USS)
could occur if some SoN (oil or dry) has been already happened.

P(FSS|oil) P(oil|FSS)
P(Oil and FSS)
P(state|finding) Prob. posterior

P(oil) P(Oil and USS)

P(USS|oil) P(oil|USS)

P(finding and state)

P(FSS|dry) P(dry|FSS)
P(Dry and FSS)
P(dry)

P(Dry and USS)

P(USS|dry) P(dry|USS)
 0,143
Oil
670
Drill 800 670

-100 -15,6 0,857

0,7 Dry
Unfavorable -130
2 0 -130
0 60

Sell
60
90 60
Do Survey
0,5
-30 123 Oil
670
Drill 800 670

-100 270 0,5

0,3 Dry
Favorable -130
1 0 -130
0 270

1 Sell
123 60
90 60

0,25
Oil
700
Drill 800 700

-100 100 0,75

Dry
No Survey -100
1 0 -100
0 100

Sell
90
90 90
 Thank You

59