Decision Trees

1/5/18
PROBLEM SOLVING:
Group 1: 3-21
Decision Analysis Group 2: 3-22

Group 3: 3-24
Group 4: 3-26
Group 5: 3-33
Group 6: 3-19 (a and b only)
Group 7: 3.28 ( a to e only)
COPYRIGHT ©2012 PEARSON EDUCATION, INC. PUBLISHING

AS PRENTICE HALL
3-2
Decision Trees 4 Key Advantages of Using

Decision Trees
Any problem that can be presented in a decision Decision trees implicitly perform variable screening
table can also be graphically represented in a or feature selection.
decision tree.
Decision trees require relatively little effort from
Decision trees are most beneficial when a sequence users for data preparation.
of decisions must be made.
Nonlinear relationships between parameters do not
All decision trees contain decision points or nodes, affect tree performance.
from which one of several alternatives may be chosen.
The best feature of using trees for analytics - easy to
All decision trees contain state-of-nature points or interpret and explain to executives!
nodes, out of which one state of nature will occur.
COPYRIGHT ©2012 PEARSON EDUCATION, INC. PUBLISHING COPYRIGHT ©2012 PEARSON EDUCATION, INC. PUBLISHING
AS PRENTICE HALL
3-3 AS PRENTICE HALL
3-4
1
1/5/18
Disadvantages of Using Five Steps of

Decision Tree Analysis
Decision Trees
1. Define the problem.
Without proper pruning or limiting tree growth, they
tend to overfit the data, making them somewhat 2. Structure or draw the decision tree.
poor predictors.
3. Assign probabilities to the states of nature.
4. Estimate payoffs for each possible combination of
alternatives and states of nature.
5. Solve the problem by computing expected
monetary values (EMVs) for each state of nature
node.
AS PRENTICE HALL
3-6
Structure of Decision Trees Thompson’s Decision Tree

A State-of-Nature Node
Trees start from left to right.
Favorable Market
Trees represent decisions and outcomes in
A Decision Node
sequential order. 1
Unfavorable Market
◦ Squares represent decision nodes.
◦ Circles represent states of nature nodes.
◦ Lines or branches connect the decisions nodes and the Favorable Market
states of nature. Construct
Small Plant
2
Unfavorable Market
Figure 3.2
AS PRENTICE HALL
3-8
2
1/5/18
Thompson’s Decision Tree EMV for Thompson Lumber

EMV for Node = (0.5)($200,000) + (0.5)(–$180,000)
1 = $10,000
Payoffs
Favorable Market (0.5) STATE OF NATURE
$200,000
Alternative with best FAVORABLE UNFAVORABLE
EMV is selected 1 ALTERNATIVE MARKET ($) MARKET ($) EMV ($)
Unfavorable Market (0.5)
–$180,000 Construct a large
200,000 –180,000 10,000
plant
Construct a small
Favorable Market (0.5) 100,000 –20,000 40,000
$100,000 plant
Construct Do nothing 0 0 0
Small Plant
2
–$20,000 Probabilities 0.50 0.50
EMV for Node = (0.5)($100,000) Table 3.9 Largest EMV

2 = $40,000 + (0.5)(–$20,000)
Figure 3.3
$0
AS PRENTICE HALL
3-10
Thompson’s Complex Decision Tree

Cost of Survey = $10,000 Thompson’s Complex Decision Tree
First Decision Second Decision Payoffs
Point Point
Favorable Market (0.78)
$190,000
1. Given favorable survey results,
2 Unfavorable Market (0.22)
–$190,000 EMV(node 2) = EMV(large plant | positive survey)
Small $90,000
Plant 3 Unfavorable Market (0.22)
–$30,000
= (0.78)($190,000) + (0.22)(–$190,000) = $106,400
No Plant
–$10,000
EMV(node 3) = EMV(small plant | positive survey)
1 Favorable Market (0.27) = (0.78)($90,000) + (0.22)(–$30,000) = $63,600
$190,000
–$190,000 EMV for no plant = –$10,000
Small
Plant 5 Unfavorable Market (0.73)
$90,000
–$30,000
2. Given negative survey results,
No Plant
–$10,000 EMV(node 4) = EMV(large plant | negative survey)
= (0.27)($190,000) + (0.73)(–$190,000) = –$87,400
$200,000 EMV(node 5) = EMV(small plant | negative survey)
–$180,000 = (0.27)($90,000) + (0.73)(–$30,000) = $2,400
Small $100,000
Figure 3.4 Plant 7 Unfavorable Market (0.50)
–$20,000 EMV for no plant = –$10,000
No Plant
$0
AS PRENTICE HALL
3-12
3
1/5/18
Thompson’s Complex Decision Tree Thompson’s Complex Decision Tree

First Decision Second Decision Payoffs
Point Point
3. Compute the expected value of the market survey, $106,400 Favorable Market (0.78)
$190,000
EMV(node 1) = EMV(conduct survey) –$190,000
$106,400
$63,600 Favorable Market (0.78)
Small $90,000
= (0.45)($106,400) + (0.55)($2,400) Plant Unfavorable Market (0.22)
–$30,000
= $47,880 + $1,320 = $49,200 No Plant
–$10,000
4. If the market survey is not conducted, –$87,400 Favorable Market (0.27)
$190,000
EMV(node 6) = EMV(large plant) $2,400
–$190,000
$2,400
Small $90,000
= (0.50)($200,000) + (0.50)(–$180,000) = $10,000 Plant Unfavorable Market (0.73)
–$30,000
EMV(node 7) = EMV(small plant) No Plant
–$10,000
$49,200
= (0.50)($100,000) + (0.50)(–$20,000) = $40,000
EMV for no plant = $0 $10,000 Favorable Market (0.50)
$200,000
5. The best choice is to seek marketing information. –$180,000
$40,000
$40,000 Favorable Market (0.50)
Small $100,000
Plant Unfavorable Market (0.50)
–$20,000
Figure 3.5
No Plant
$0
AS PRENTICE HALL
3-14
Expected Value of Sample Information Sensitivity Analysis

Suppose Thompson wants to know the actual value of doing the survey. n How sensitive are the decisions to changes
in the probabilities?
Expected value Expected value n How sensitive is our decision to the probability
with sample
EVSI = information, assuming –
of best decision of a favorable survey result?
without sample
no cost to gather it information n That is, if the probability of a favorable result (p
= .45) where to change, would we make the
= (EV with sample information + cost) same decision?
– (EV without sample information) n How much could it change before we would
make a different decision?
EVSI = ($49,200 + $10,000) – $40,000 = $19,200
AS PRENTICE HALL
3-16
4
1/5/18
Sensitivity Analysis Bayesian Analysis
p = probability of a favorable survey result

n There are many ways of getting
(1 – p) = probability of a negative survey result probability data. It can be based on:
EMV(node 1) = ($106,400)p +($2,400)(1 – p) n Management’s experience and intuition.
= $104,000p + $2,400 n Historical data.
We are indifferent when the EMV of node 1 is the n Computed from other data using Bayes’
same as the EMV of not conducting the survey, theorem.
$40,000
n Bayes’ theorem incorporates initial
$104,000p + $2,400 = $40,000 estimates and information about the
$104,000p = $37,600 accuracy of the sources.
p = $37,600/$104,000 = 0.36 n It also allows the revision of initial
If p<0.36, do not conduct the survey. If p>0.36, estimates based on new information.
conduct the survey.
AS PRENTICE HALL
3-18
DECISION ANALYSIS IN ACTION:

Utility Theory Ford Motor Company
Monetary value is not always a true indicator of Manufactures about 5M cars and trucks annually
the overall value of the result of a decision. Employs more than 200,000 people at about 100
The overall value of a decision is called utility. facilities around the globe.
Economists assume that rational people make Large supplier decisions under tight deadlines.
decisions to maximize their utility. Input data about their suppliers as well as the
type of decision criterion they want to use.
Utility assessment assigns the worst outcome a utility of 0,
and the best outcome, a utility of 1. Model outputs the best set of suppliers to meet
the specified needs.
A standard gamble is used to determine utility values.
Savings of over $40 million annually.
When you are indifferent, your utility values are equal.
AS PRENTICE HALL
3-20
5
1/5/18
DECISION ANALYSIS IN ACTION: PROBLEM SOLVING:

Gerber Medical Clinic
Decide whether to continue using PVC in their A group of medical professionals is considering the construction of a
baby products private clinic. If the medical demand is high (i.e., there is a favorable
market for the clinic), the physicians could realize a net profit of
Greenpeace claims that PVC is dangerous $100,000. If the market is not favorable, they could lose $40,000. Of
course, they don’t have to proceed at all, in which case there is no
A month before Christmas, the US CPSC plan to cost.
issue a press release advising dangers.
Choice: reactive or proactive In the absence of any market data, the best the physicians can guess
Using a decision tree, Gerber decided to be is that there is a 50–50 chance the clinic will be successful.
proactive and initiate its own solutions and Construct a decision tree to help analyze this problem. What should
hoping for a favorable report. the medical professionals do?
AS PRENTICE HALL
3-22
PROBLEM SOLVING: PROBLEM SOLVING:

Medical Clinic Medical Clinic 2
The physicians in have been approached by a market research firm that offers to perform
EMV for node 1 = 0.5(100,000) + 0.5(–40,000) = $30,000. a study of the market at a fee of $5,000. The market researchers claim their experience
enables them to use Bayes’ theorem to make the following statements of probability:
Choose the highest EMV, therefore construct the clinic.
probability of a favorable market given a favorable study = 0.82

probability of an unfavorable market given a favorable study = 0.18
probability of a favorable market given an unfavorable study = 0.11
probability of an unfavorable market given an unfavorable study = 0.89
probability of a favorable research study = 0.55
probability of an unfavorable research study = 0.45
A. Develop a new decision tree for the medical professionals to reflect the options now
open with the market study.
B. Use the EMV approach to recommend a strategy.
C. What is the expected value of sample information? How much might the physicians be
willing to pay for a market study?
AS PRENTICE HALL
3-24
6
1/5/18
PROBLEM SOLVING: PROBLEM SOLVING:

Medical Clinic 2 Medical Clinic 2
b. EMV(node 2) = (0.82)($95,000) + (0.18)(–$45,000)
= 77,900 – 8,100 = $69,800
EMV(node 3) = (0.11)($95,000) + (0.89)(–$45,000)
= 10,450 – $40,050 = –$29,600
EMV(node 4) = $30,000
EMV(node 1) = (0.55)($69,800) + (0.45)(–$5,000)
= 38,390 – 2,250 = $36,140
The EMV for using the survey = $36,140.
EMV(no survey) = (0.5)($100,000) + (0.5)(–$40,000)
= $30,000
The survey should be used..
AS PRENTICE HALL
3-26
PROBLEM SOLVING:
Medical Clinic 2
c. EVSI = ($36,140 + $5,000) – $30,000 = $11,140.

Thus, the physicians would pay up to $11,140 for the
survey.
COPYRIGHT ©2012 PEARSON EDUCATION, INC. PUBLISHING

AS PRENTICE HALL
3-27

Decision Trees

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Decision Trees

Uploaded by

Copyright:

Available Formats

1/5/18

Decision Analysis Group 2: 3-22

COPYRIGHT ©2012 PEARSON EDUCATION, INC. PUBLISHING

Decision Trees 4 Key Advantages of Using

Disadvantages of Using Five Steps of

Structure of Decision Trees Thompson’s Decision Tree

Thompson’s Decision Tree EMV for Thompson Lumber

EMV for Node = (0.5)($100,000) Table 3.9 Largest EMV

Thompson’s Complex Decision Tree

Thompson’s Complex Decision Tree Thompson’s Complex Decision Tree

Expected Value of Sample Information Sensitivity Analysis

EVSI = ($49,200 + $10,000) – $40,000 = $19,200

Sensitivity Analysis Bayesian Analysis

p = probability of a favorable survey result

DECISION ANALYSIS IN ACTION:

DECISION ANALYSIS IN ACTION: PROBLEM SOLVING:

PROBLEM SOLVING: PROBLEM SOLVING:

probability of a favorable market given a favorable study = 0.82

PROBLEM SOLVING: PROBLEM SOLVING:

b. EMV(node 2) = (0.82)($95,000) + (0.18)(–$45,000)

= 77,900 – 8,100 = $69,800

EMV(node 3) = (0.11)($95,000) + (0.89)(–$45,000)

= 10,450 – $40,050 = –$29,600

EMV(node 1) = (0.55)($69,800) + (0.45)(–$5,000)

= 38,390 – 2,250 = $36,140

The EMV for using the survey = $36,140.

EMV(no survey) = (0.5)($100,000) + (0.5)(–$40,000)

The survey should be used..

c. EVSI = ($36,140 + $5,000) – $30,000 = $11,140.

COPYRIGHT ©2012 PEARSON EDUCATION, INC. PUBLISHING

You might also like