TOPSIS

• Technique
h i ffor O
Order
d Preference
f b
by Si
Similarity
il i to
Ideal Solution
• Yoon and Hwang introduced the TOPSIS method
based on the idea that the best alternative should
have the shortest distance from the positive ideal
solution and farthest distance from the negative
id l solution.
ideal l i
• (K. Paul Yoon and Ching‐Lai Hwang, “Multiple
Attribute Decision Making: An Introduction”, Sage
Publications, USA, 1995).

Decision Analysis 2009 Fall ‐ Dr. Bahar Sennaroğlu

 The steps of TOPSIS method (1/6)
• Step 1. Calculate the normalized decision
matrix R =  rij  mxn . The normalized value rij is
calculated as
xij
rij = i = 1, 2,..., m; j = 1, 2,..., n
m

∑ ij
x 2

i =1
Th number
The b off alternatives
lt ti Th number
The b off attributes
tt ib t

• Th
The normalization
li ti is
i ddone for
f convenience
i off
comparison by converting different units of
attributes to an unified unit.
Decision Analysis 2009 Fall ‐ Dr. Bahar Sennaroğlu
The steps of TOPSIS method (2/6)
• Step 2. Calculate the weighted normalized
decision matrix V = vij  mxn . The weighted
g
normalized value vij is calculated as
vij = ( w j )( rij ) i = 1, 2,..., m; j = 1, 2,..., n

where wj is the weight of the jth attribute and

n

∑w
j =1
j =1

Decision Analysis 2009 Fall ‐ Dr. Bahar Sennaroğlu

The steps of TOPSIS method (3/6)
• Step 3. Determine the positive ideal solution
((PIS)) A+ and negative
g ideal solution ((NIS)) A‐.
A+ ={( i
)( i
) }{ }
max vij | j ∈ J , min vij | j ∈ J ′ , i = 1, 2,..., m = v1+ , v2+ ,..., vn+

A = {( min v | j ∈ J ) , ( max v | j ∈ J ′ ) , i = 1, 2,..., m} = {v , v ,..., v }

−
ij ij
−
1
−
2
−
n
i i

where J is a set of benefit attributes and J’ is a

sett off costt attributes.
tt ib t

Decision Analysis 2009 Fall ‐ Dr. Bahar Sennaroğlu

The steps of TOPSIS method (4/6)
• Step 4. Calculate the separation measures, using the
n‐dimensional Euclidean distance.
• The separation of each alternative from the positive
ideal solution
n
Si+ = ∑ ijj j )
(
j=1
v − v + 2
i = 1, 2,..., m
• The separation of each alternative from the negative
ideal solution
n
Si− = ∑ ij j )
( v
j =1
− v − 2
i = 1,
1 22,..., m
Decision Analysis 2009 Fall ‐ Dr. Bahar Sennaroğlu
The steps of TOPSIS method (5/6)
• Step 5. Calculate the relative closeness to the
ideal solution.

Si−
Ci = + i = 1, 2,..., m; 0 ≤ Ci ≤ 1
Si + Si−

Decision Analysis 2009 Fall ‐ Dr. Bahar Sennaroğlu

The steps of TOPSIS method (6/6)
• Step 6. Rank the alternatives with respect to Ci
in decendingg order.
• The preferred alternative should have the
shortest distance from the positive ideal
solution and the farthest distance from the
negative ideall solution,
l where
h a higher
h h Ci
would mean higher preference.

Decision Analysis 2009 Fall ‐ Dr. Bahar Sennaroğlu

Example: Choosing an Automobile
• SSuppose you are bbuying
i a car, and
d you are iinterested
d
in both price and life span.
• Two
T conflicting
fli i objectives:
bj i
– A long expected life span
– A low
l price
i
• Three alternatives:
– The
h Portalol (a
( relatively
l i l expensive
i sedand with
i h a reputation
i
for longevity)
– The Norushi (renowed for its reliability)
– The Standart Motors car (a relatively inexpensive domestic
automobile)

Decision Analysis 2009 Fall ‐ Dr. Bahar Sennaroğlu

Example: Choosing an Automobile
Attribute Portalo Norushi Standard
Price ($1000s) 17 10 8
Life Span (Years) 12 9 6

k P = 0.714
0 714
k L = 0.286

Decision Analysis 2009 Fall ‐ Dr. Bahar Sennaroğlu

 The normalized decision matrix
Alternatives ( i =1,2,3) P L
Attributes ( j =11,2)
2) Portalo Norushi Standard 17 12 
Price ($1000s) 17 10 8 X = 10 9 
Life Span (Years) 12 9 6  8 6 
xij
rij = i = 1, 2,..., m; j = 1, 2,..., n
R =  rij  m
mxn ∑ ij
x 2

i =1

 17 12   17 12 
   
 17 + 10 + 8 12 + 9 + 6  
2 2 2 2 2 2
453 261 
 10 9   10 9 
R= = 
 17 + 10 + 8
2 2 2
12 + 9 + 6  
2 2 2
453 261 
 8 6   8 6 
  
 17 2 + 102 + 82 12 + 9 + 6  
2 2 2 453 261 
Decision Analysis 2009 Fall ‐ Dr. Bahar Sennaroğlu
 The weighted normalized decision matrix
k P = w1 = 0.714
V = vij  k L = w2 = 0.286
mxn n

∑w
j =1
j =1

vij = ( w j )( rij ) i = 1, 2 m; j = 1,
1 2,..., 1 2,...,
2 n

 17 12 
(0.714)
(0 714) (0
(0.286)
286) 
 453 261  0.5703 0.2124
 
 10 9   
V = (0.714)
(0 714) (0286) = 0
0.3355
3355 0.1593
0 1593
 453 261   
 0.2684 0.1062 
 8 6 
(0.714)
(0 714) (0
(0.286)
286)
261 
 453
Decision Analysis 2009 Fall ‐ Dr. Bahar Sennaroğlu
 The positive ideal solution A+ and
negative ideal solution A‐
A+ ={( max v | j ∈ J ) , ( min v | j ∈ J ′) , i = 1, 2,..., m} = {v , v ,..., v }
i
ij
i
ij
+
1
+
2
+
n

A = {( min v | j ∈ J ) , ( max v | j ∈ J ′ ) , i = 1,
−
ij ij 2 m} = {v , v ,..., v }
1 2,..., −
1
−
2
−
n
i i

where J is a set of benefit attributes and J’ is a set of cost attributes.

P L
 0.5703 0.2124 
V =  0.3355 0.1593
 0.2684 0.1062 
A+ = {0.2684, 0.2124}
A− = {0.5703, 1062}
0 5703 00.1062

Decision Analysis 2009 Fall ‐ Dr. Bahar Sennaroğlu

 The separation of each alternative from A+ and A‐
n
 0.5703 0.2124 
S =
i
+
∑ (vj =1
ij −v )+ 2
j i = 1, 2,..., m
V =  0.3355 0.1593
n  0.2684 0.1062 
Si− = ∑ ij j )
(
j=1
v − v − 2
i = 1, 2,..., m
{ }
A+ = v1+ , v2+ = {0.2684, 2124}
0 2684 00.2124
A− = {v , v } = {0.5703, 0.1062}
−
1
−
2

Separation Measures
Alternatives
Si+ Si−
Portalo (0.5703-0.2684)2 + (0.2124-0.2124)2 = (0.5703-0.5703)2 + (0.2124-0.1062)2 =
0.3019 0.1062
N
Norushi
hi 3355 0 2684)2 + (0.1593-0.2124)
(0.3355-0.2684)
(0 (0 1593 0 2124)2 = 3355 0 5703)2+ (0.1593-0.1062)
(0.3355-0.5703)
(0 (0 1593 0 1062)2 =
0.0856 0.2408
Standard (0.2684-0.2684)2 + (0.1062-0.2124)2 = (0.2684-0.5703)2 + (0.1062-0.1062)2 =
0.1062 0.3019

Decision Analysis 2009 Fall ‐ Dr. Bahar Sennaroğlu

 The relative closeness to the ideal solution and
rank of the alternatives with respect to Ci in decending order
−
S
Ci = + i
i = 1, 2,..., m; 0 ≤ Ci ≤ 1
Si + Si−

Alternatives Si+ Si− Ci Rank

Portalo 0.3019 0.1062 0.2602 3

Norushi 0 0856
0.0856 0 2408
0.2408 0 7378
0.7378 2

Standard 0.1062 0.3019 0.7398 1

The preferred alternative is the Standard when weights are

assessed as 0.714 for price and 0.286 for the expected life span.
Decision Analysis 2009 Fall ‐ Dr. Bahar Sennaroğlu