Multi Criteria Decision Making

(MCDM)

1
• What is Multi Criteria Decision Making?
• Difficulties in solving MCDM problems
• Classification of MCDM problems
• Multi Criteria Selection Problems
• Multi Criteria Mathematical Programming

2
• Most optimization models are based on a single
criterion; minimize cost or maximize profit
• However, decision making under multiple criteria,
generally conflicting, is more common in
practice.
• How to compare “apples”(objective 1) and
(objective 2) “oranges”?
• In MCDM, alternatives are measured by several
conflicting criteria and the problem is to choose
the “best” alternative.
3
• No Optimal Solution
• Not possible to compare two solutions objectively
• Need to involve the Decision Maker (DM) ----
One of the rare areas of Operations Research,
where the DM plays a key role during the
solution process!

4
 MCDM Classification
• MCDM Problems can be broadly classified into
two types.
✓ Multi Criteria Selection Problems (MCSP)
✓ Multi Criteria Mathematical Programming
(MCMP) problems

5
 1. Multi Criteria Selection Problems
(MCSP)
– Finite number of alternatives
– Alternatives measured by conflicting criteria
– Rank the alternatives from best to worst

Applications
– Buying a car, house, computer, etc.
– Ranking colleges (e.g: U.S. News)
– Supplier Selection Problems
– Bid Selection Problems

6
 2. Multi Criteria Mathematical
Programming (MCMP) Problems
– Alternatives are defined by explicit constraints,
resulting in infinite number of alternatives
– Several conflicting objective functions
– Determine the Best Compromise Solution

MCMP Applications
– Portfolio Optimization
– Quality Control
– Health Planning
– Project Management
– Supply Chain Network Design
7
 Multi Criteria Ranking Problems

• Dominated and Nondominated Alternatives.

• Ideal solution: Vector of maximum values of each
objective, ignoring other objectives. 8
Faculty Recruiting (finite alternatives)

Ideal Solution (8,8,5) 9

 Definitions
• Ideal Solution: Best values achievable for each
objective ignoring other objectives.

• Efficient Solution: Can improve an objective only

at the expense of at least one other objective,
e.g., Candidate 1

• Dominated Solution: Possible to improve an

objective without losing achievements in other
objectives, e.g., Candidate 2

10
 Multi Criteria Ranking methods:
Rating (Scoring) Method
– Each criterion rated in a scale of 1 to 10
– Normalize to obtain weights
– Use weighted sum of criteria to rank
• Let Vj = Value/Rating assigned to criterion j (j = 1,...,P)

Assume that the ratings for research, teaching and service are 9, 7 and
4 respectively.

11
 Weights by ranking criteria(Borda
method)
• Named after Jean Charles de Borda, 18th Century French
Physicist
• The P criteria are ranked 1(most important) to P (least
Important)
– Criterion ranked 1 gets P points, 2nd rank gets P-1
points, …., last place 1 point.
– Let S = sum of all points = P(P+1)/2
• Weights
– criterion 1 = P/S
– criterion 2 = (P-1)/S
– last criterion= 1/S
Assume Rank 1 = Research, so 3 points.
Rank 2 = Teaching, so 2 points.
Rank 3 = Service, so 1 point.
12
 Paired Comparison Method
• Ask DM for preference information between
pairs of criteria.
e.g. Between criteria A and B, DM can
respond
✓ A is preferred to B
✓ B is preferred to A
✓ Indifferent
• N(N-1)/2 Comparisons between pairs of criteria
13
 Paired Comparison Method
Example
Five criteria A, B, C, D, E. Requires 10 paired
comparisons
• A>B A>C A>D A>E
• B<C B>D B<E
• C>D C<E
• D<E
• Assign numerical values for preferences: 1 if
preferred or 0 if not preferred. (Indifference -
assign "1" to both criteria)

14
Construct pair-wise comparison matrix
Step 1: Construct the pairwise comparison matrix P:

P(5x5) A B C D E

A 1 1 1 1 1

B 0 1 0 1 0

C 0 1 1 1 0

D 0 0 0 1 0

E 0 1 1 1 1

Step 2: Compute the row sum as A is 5; B is 2; C is 3; D is 1; E is 4

15
 Paired Comparison Method
Step 3: The ranking of the five crieteria A, E, C, B, D
• Note: Weights are
wA = 5/15= 0.333
wB = 2/15 = 0.133
wC = 3/15 = 0.2
wD = 1/15 = 0.067
wE = 4/15 = 0.267
• Consistency Problem: Inconsistent response
causing Intransitivity, e.g., A>B, B>C; then DM
says C>A
16
 Issues in using Ranking Methods
• Consistency of the Decision Maker
• Scaling of criteria values
– If criteria values are not scaled properly,
larger criteria values will skew the results
in their favor, irrespective of the chosen
weights
– Several scaling methods are available

17
 Multiple Criteria Mathematical
Programming(MCMP) Problem
• MAXIMIZE F (x)= {f1(x), f2(x), ..., fk(x)}
• Subject to gj(x) <= 0 j = 1,...,M
• Where x is an n-vector of decision variables
and fi(x), i=1,...k are the K Criteria functions.
• Let S = { x | gj(x) <= 0, for all j } =
S is called the Decision space
• Let Y = { y | F(x) = y for some x ε S} =
Y is called the Objective space
18
Decision Space vs. Objective Space

19
EFFICIENT/NON-DOMINATED/PARETO
OPTIMAL SOLUTION
• A solution x0 ε S is said to be efficient if fk(x) >
fk(x0) for some x ε S implies that fj(x) < fj(x0)
for at least one other index j.
• More simply stated, an efficient solution has
the property that an improvement in any one
criterion is possible only at the expense of at
least one other criterion.

20
EFFICIENT SET:
• Set of all efficient solutions is called the efficient set.
• Solution of MCMP reduces to determining the
efficient set (but not practical).
ASSUMPTION:
• DM has "preferences" among the efficient solutions.
A best compromise solution is an efficient solution
that is 'best' with respect to the DM's
preference/utility function.
MCMP PROBLEM:
Determine the best compromise solution.

21
Pareto-Optimality Example

22
23
24
25
 Illustration
• Ideal Solution; Point (26,15) but infeasible
• Dominated solution; Points A, B and E
• Non-dominated or efficient or Pareto optimal
solution; Points C and D
• North-East rule to identify efficient solutions
when both objectives are Max
• Efficient set; all points on the line (C,D)
• Bounds on the objectives:
– For Z1 : (18,26); For Z2 : (9,15)

26
 Classification of MCMP Methods
1. When complete information about the
preference function is available from DM,
then Goal Programming is used.
2. When no information is available, then
compromising programming is used.
3. Where partial information is obtainable
progressively from DM, then interactive
methods are used.

27