International Journal of Research in Engineering and Science (IJRES)

ISSN (Online): 2320-9364, ISSN (Print): 2320-9356

www.ijres.org Volume 1 Issue 2 ǁ June. 2013 ǁ PP.05-12

Notes on TOPSIS Method

Zlatko Pavić1, Vedran Novoselac2
1,2
(Mechanical Engineering Faculty in Slavonski Brod, University of Osijek, Croatia)

ABSTRACT: The paper demonstrates the application of TOPSIS method using two selected examples. In the
first example, it is shown that the best TOPSIS solution is neither closest to the positive ideal solution nor the
farthest from the negative ideal solution. In many works on TOPSIS method stands as follows: "The basic
principle is that the chosen alternative should have the shortest distance from the positive ideal solution and the
longest distance from the negative ideal solution".
Keywords: decision matrix, negative ideal solution, positive ideal solution

I. INTRODUCTION
Multi-criteria decision making.
In a general sense, it is the aspiration of human being to make "calculated" decision in a position of
multiple selection. In scientific terms, it is the intention to develop analytical and numerical methods that take into
account multiple alternatives with multiple criteria.
TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) is one of the numerical
methods of the multi-criteria decision making. This is a broadly applicable method with a simple mathematical
model. Furthermore, relying on computer support, it is very suitable practical method. The method is applied in
the last three decades (on the history of TOPSIS see [4], [3]), and there are many papers on its applications (see
[11], [8], [9]).

Description of the problem.

Given m options (alternatives) Ai , each of which depends on n parameters (criteria) x j whose
values are expressed with positive real numbers xij . The best option should be selected.

Mathematical model of the problem.

Initially, the parameter values xij should be balanced according to the procedure of normalization.
Suppose that aij are the normalized parameter values. Then each option Ai is expressed as the point

Ai (ai1 , , ain )  R n . Selecting the most optimal value aj {a1 j , , amj } for every parameter x j , we
 
determine the positive ideal solution A = ( a1 , , an ) . The opposite is the negative ideal solution
A = (a1 , , an ) . The positive and negative ideal solution are also denoted by A and A . The decision on
the order of options is made respecting the order of numbers
d ( Ai , A ) 1
Di = = . (1)
d ( Ai , A )  d ( Ai , A ) d ( Ai , A ) / d ( Ai , A )  1
  

The option Ai is the best solution if max{D1 , D2 , , Dm } = Di , and the option Ai is the worst solution
1 1 2

if min{D1 , D2 , , Dm } = Di . The other options are between these two extremes. The maximum distance
2

D = max i =1, ,mDi is usually called TOPSIS metric.

Geometrical image of the problem.

Fig. 1 shows the initial arrangement of alternatives in TOPSIS method for n = 2 . Parameter x1 = x1

has a monotonically increasing preference, and parameter x2 = x2 has a monotonically decreasing preference.
The positive A and negative A ideal solution are located at diagonally opposite positions. The best solution

www.ijres.org 5 | Page
 Notes on TOPSIS Method

is the alternative A7 .

Figure 1. Geometrical representation of TOPSIS method

TOPSIS is a compensatory method. These kinds of methods allow the compromise between different
criteria, where a bad result in one criterion can be compensated by a good result in another criterion. An
assumption of TOPSIS method is that each criterion has either a monotonically increasing or decreasing
preference. Due to the possibility of criteria modelling, compensatory methods, certainly including TOPSIS, are
widely used in various sectors of multi-criteria decision making (see [10], [2], [1]).

II. COMPUTATIONAL PROCEDURE FOR TOPSIS METHOD

Problem.
We examine m alternatives A1 , , Am . Each alternative Ai respects n criteria x1 , , xn which are
expressed with positive numbers xij . The criteria x1 , , xk are benefit (monotonically increasing preference),
and criteria xk 1 , , xn are non-benefit (monotonically decreasing preference). Weights w j of the criteria x j


n
are given so that j =1
w j = 1 . It is necessary to select the most optimal alternative.
Initial Table and Decision Matrix.
For better visibility, the given alternatives, criteria and its weights are placed in the table (see Table 1).

Table 1. Initial table for TOPSIS method

www.ijres.org 6 | Page
 Notes on TOPSIS Method

The given numbers xij and their matrix

 x11 x12 x1n 

x x22 x2 n 
 21
X =  (2)
 
 xm1 xm 2 xmn 
 
must be balanced, since the numbers xij present values of different criteria with different measuring units. One
must also take into account the given weights w j of the criteria x j . First, the measuring numbers xij of the
criteria x j are replaced with the normalized or relative numbers
xij
rij = (3)
m

x
i =1
2
ij

belonging to the open interval 0,1 . Then, according to the share w j x j of the criteria x j , the normalized
numbers rij are replaced with the weighted normalized numbers
xij
aij = w j rij = w j (4)
m

x
i =1
2
ij

belonging to 0,1 . The further data processing uses the weighted normalized decision matrix
 a11 a12 a1n 
a a2 n 
 21 a22
A = . (5)
 
 am1 am 2 amn 
 
If all weights w j are mutually equal, in which case w j = 1 / n , the numbers rij can be applied in the matrix
A as the numbers aij .

Working Table.
The weighted normalized decision matrix A and all the data that will be calculated, we try to write in
one table.

Table 2. Working table for TOPSIS method

www.ijres.org 7 | Page
 Notes on TOPSIS Method

The coordinates a j of the positive ideal solution A = (a1 a2 an ) are chosen using the formula

maxaij for j = 1, ,k
 i
aj =  minaij for j = k  1, , n. (6)
 i


If some alternative Ai is equal to A , then it is obvious that the alternative Ai is the best solution. If it is not,
0 0
then we continue the procedure.

The coordinates a j of the negative ideal solution A = (a1 a2 an ) are chosen applying the
formula

 minaij for j = 1, ,k

 i
a j = maxaij for j = k  1, , n. (7)
 i


   
The numbers d i of the column d = (d1 d 2 d m ) are the distances from the points Ai to the
point A , which is calculated by the formula

a  aj  .
n 2
di = d ( Ai , A ) = ij (8)
j =1

   
The numbers d i of the column d = ( d1 d 2 d m )  are the distances from the points Ai to the
point A , which is calculated by the formula

a  a j  .
n 2
di = d ( Ai , A ) = ij (9)
j =1

   
The numbers Di of the column D = ( D1 D2 Dm )  are the relative distances of the points Ai
respecting the points A and A , which is expressed by the formula

di d ( Ai , A )
Di = = . (10)
di  di d ( Ai , A )  d ( Ai , A )

If max{D1 , D2 , , Dm } = Di , then we accept the alternative Ai as the best solution. If
1 1

min{D1 , D2 , , Dm } = Di , then we accept the alternative Ai as the worst solution.
2 2

www.ijres.org 8 | Page
 Notes on TOPSIS Method

III. TWO EXAMPLES OF USING TOPSIS METHOD

Example 1. Four alternatives with three criteria are given in Table 3. The criteria x1 and x 2 are benefit, and the
criterion x3 is non-benefit. The weights of the criteria are equal. Decide which alternative is the best.

Table 3. Initial table for Example 1

Since the weights are equal, we can use the normalized decision matrix A with the elements
xij
aij = (11)
4

x
i =1
2
ij

for i = 1, 2,3, 4 and j = 1, 2,3 . In this case, the matrix A reads as follows:
0,535 0, 441 0, 407 
0, 688 0, 441 0, 488 
 
A = 0,382 0,515 0, 732 . (12)
 
0,306 0,588 0, 244 
 

Relying on the matrix A , we have to determine two rows ( A , A ) and three columns ( d  , d  , D ) in the
working table.

Table 4. Working table for Example 1

   
According to the formula in (6), the positive ideal solution A = ( a1 a2 a3 ) contains the greatest
numbers of the first and second column of A , and the smallest number of the third column of A .
   
According to the formula in (7), the negative ideal solution A = (a1 a2 a3 ) contains the smallest
A , and the greatest number of the third column of A .
numbers of the first and second column of
      
The distances d = (d d d d ) , from the alternatives Ai to the positive ideal solution A , are calculated
1 2 3 4

www.ijres.org 9 | Page
 Notes on TOPSIS Method

by the formula in (8) with n = 3 , so

a  aj  .
3 2
di = d ( Ai , A ) = ij (13)
j =1
     
The distances d = ( d1 d 2 d 3 d 4 ) , from the alternatives Ai to the negative ideal solution A , are
calculated by the formula in (9) with n = 3 , so

a  a j  .
3 2
di = d ( Ai , A ) = ij (14)
j =1

     
The relative distances D = ( D1 D2 D3 D4 ) of the alternatives Ai respecting the positive ideal
 
A and negative ideal solution A are calculated using the formula in (10), so
solution
d
Di =  i  . (15)
di  di

Applying the last tree columns of the Table 4 we have the following three preferred orders of alternatives:
A2 A1 A4 A3 by the column of D from the largest to smallest number
A4 A2 A1 A3 by the column of d  from the largest to smallest number
A1 A2 A4 A3 by the column of d  from the smallest to largest number
TOPSIS method prefers the first order respecting the column of D .

In the following example we have the most desirable combination: d , d

 
and D point to one and
the same best solution.

Example 2. 1Six alternatives with five criteria and their weights are given in Table 5. The criteria x1 , x2 , x3 are
benefit, and the criteria x4 , x5 are non-benefit. Select the best alternative.

Table 5. Initial table for Example 2

Weighted normalized decision matrix A with the elements

xij
aij = w j (16)
6

x
i =1
2
ij

for i = 1, ,6 and j = 1, ,5 stands as follows:

www.ijres.org 10 | Page
 Notes on TOPSIS Method

 0,118 0, 049 0, 057 0,118 0, 033

 0,158 0, 035 0, 043 0,103 0, 088

0, 097 0, 028 0,113 0, 044 0, 077 
A = . (17)
 0, 079 0, 063 0, 099 0, 088 0, 055
 0,178 0, 028 0, 071 0,103 0, 066 
 
 0, 059 0, 028 0, 085 0,133 0, 022 

     
The positive ideal solution A = ( a1 a2 a3 a4 a5 ) contains the greatest numbers of the first, second and
third column of A , and the smallest numbers of the fourth and fifth column of A .
     
The negative ideal solution A = (a1 a2 a3 a4 a5 ) contains the smallest numbers of the first, second
and third column of A , and the greatest numbers of the fourth and fifth column of A .
        
The distances d = ( d1 d 2 d 3 d 4 d 5 d 6 ) , from the alternatives Ai to the positive ideal solution A ,
are calculated applying the distance formulas in (8) with n = 5 .
       
The distances d = ( d1 d 2 d 3 d 4 d 5 d 6 ) , from the alternatives Ai to the negative ideal solution
A , are calculated applying the distance formulas in (9) with n = 5 .
       
The relative distances D = ( D1 D2 D3 D4 D5 D6 ) of the alternatives Ai respecting the positive
ideal solution A and negative ideal solution A are determined using the quotient formulas in (10).

Table 6. Working table for Example 2

Applying the last tree columns of the Table 6 we have the following three preferred orders of
alternatives:

A5 A3 A2 A4 A1 A6 by the column of D from the largest to smallest number

A5 A3 A2 A4 A1 A6 by the column of d  from the largest to smallest number
A5 A3 A1 A4 A2 A6 by the column of d  from the smallest to largest number

The first order is preferred by TOPSIS method.

www.ijres.org 11 | Page
 Notes on TOPSIS Method

IV. CONCLUSION
In the method presenting it is important to find examples that adequately show its meaning and
application. After that, the generalizations of the method can be implemented to extend its applications. In the case
of TOPSIS method the first generalization refers to the processing of insufficiently precise data, namely, fuzzy
data (see [6], [5], [1]).
The second generalization applies to the norm and metric. Let p  1 be a real number. Using the p
-norm in the normalization procedure, we get
xij xij
rij = = . (18)
|| ( x1 j , , xmj ) || p m
p
| x
i =1
ij | p

Then, using the p -metric in the distance calculation, we have

n
di = d p ( Ai , A ) = p
| a
j =1
ij  aj | p . (19)

The max-norm and max-metric can also be applied in the computational procedure of TOPSIS method .

REFERENCES
[1] A. Balin, P. Alcan, and H. Basligil, Co performance comparison on CCHP systems using different fuzzy
multi criteria decision making models for energy sources, Fuelling the Future, pp. 591-595, 2012.
[2] I. B. Huang, J. Keisler, and I. Linkov, Multi-criteria decision analysis in environmental science: ten years of
applications and trends, Science of the Total Environment 409, pp. 3578-3594, 2011.
[3] C. L. Hwang, Y. J. Lai, and T. Y. Liu, A new approach for multiple objective decision making, Computers
and Operational Research 20, pp. 889-899, 1983.
[4] C. L. Hwang, and K. Yoon, Multiple Attribute Decision Making: Methods and Applications, Berlin
Heidelberg New York, Springer-Verlag, 1981.
[5] G. R. Jahanshahloo, F. Hosseinzadeh Lotfi, and M. Izadikhah, Extension of the TOPSIS method for
decision-making problems with fuzzy data, Applied Mathematics and Computation, pp. 1544-1551, 2006.
[6] Y. J. Lai, T. Y. Liu, and C. L. Hwang, Fuzzy Mathematical Programming: Methods and Applications, Berlin
Heidelberg New York, Springer-Verlag, 1992.
[7] Y. J. Lai, T. Y. Liu, andC. L. Hwang, TOPSIS for MODM, European Journal of Operational Research 76,
pp. 486-500, 1994.
[8] G. H. Tzeng, and J. J. Huang, Multiple Attribute Decision Making: Methods and Applications, New York,
CRC Press, 2011.
[9] J. Xu, and Z. Tao, Rough Multiple Objective Decision Making, New York, CRC Press, 2012.
[10] K. A. Yoon, A reconciliation among discrete compromise situations, Journal of Operational Research
Society 38, pp. 277-286, 1987.
[11] K. P. Yoon, and C. Hwang, Multiple Attribute Decision Making: An Introduction, California, SAGE
Publications, 1995.
[12] E. K. Zavadskas, A. Zakarevicius, and J. Antucheviciene, Evaluation of ranking accuracy in multi-criteria
decisions, Informatica 17, pp. 601-618, 2006.

www.ijres.org 12 | Page