roman słowiński poznań university of technology, poland

25
Multiple-criteria ranking using an additive utility function constructed via ordinal regresion : UTA method Roman Słowiński Poznań University of Technology, Poland Roman Słowiński

Upload: palmer-malone

Post on 03-Jan-2016

42 views

Category:

Documents


0 download

DESCRIPTION

Multiple-criteria ranking using an additive utility function constructed via ordinal regresion : UTA method. Roman Słowiński Poznań University of Technology, Poland.  Roman Słowiński. g 2 ( x ). g 2max. A. g 2min. g 1 ( x ). g 1min. g 1max. Problem statement. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Roman Słowiński Poznań University of Technology, Poland

Multiple-criteria ranking

using an additive utility function

constructed via ordinal regresion :

UTA method

Roman SłowińskiPoznań University of Technology, Poland

Roman Słowiński

Page 2: Roman Słowiński Poznań University of Technology, Poland

2

Problem statement

Consider a finite set A of alternatives (actions, solutions)

evaluated by n criteria from a consistent family G={g1,...,gn}

g2max

g1(x)

g2(x)

g2min

g1min g1max

A

Page 3: Roman Słowiński Poznań University of Technology, Poland

3

Problem statement

Consider a finite set A of alternatives (actions, solutions)

evaluated by n criteria from a consistent family G={g1,...,gn}

g2(x)

g1(x)

A

Page 4: Roman Słowiński Poznań University of Technology, Poland

4

Problem statement

Taking into account preferences of a Decision Maker (DM),

rank all the alternatives of set A from the best to the worst

A

**

**

x

xx

xx

x

*

*

x

x * *

x x * *

x x x * * x

x x x

x

Page 5: Roman Słowiński Poznań University of Technology, Poland

5

Basic concepts and notation

Xi – domain of criterion gi (Xi is finite or countably infinte)

– evaluation space

x,yX – profiles of alternatives in evaluation space

– weak preference (outranking) relation on X: for each x,yX

xy „x is at least as good as y”

xy [xy and not yx] „x is preferred to y”

x~y [xy and yx] „x is indifferent to y”

n

iiXX

1

Page 6: Roman Słowiński Poznań University of Technology, Poland

6

n

iii aguaU

1

For simplicity: Xi , for all i=1,…,n

For each gi, Xi=[i, i] is the criterion evaluation scale, i i , where

i and i, are the worst and the best (finite) evaluations, resp.

Thus, A is a finite subset of X and

where g(a) is the vector of evaluations of alternative aA on n criteria

Additive value (or utility) function on X: for each aX

where ui are non-decreasing marginal value functions, ui : Xi ,

i=1,...,n

Basic concepts and notation

n

iiiiii ,aa;,A:g

1

g

Page 7: Roman Słowiński Poznań University of Technology, Poland

7

Criteria aggregation model = preference model

To solve a multicriteria decision problem one needs

a criteria aggregation model, i.e. a preference model

Traditional aggregation paradigm:

The criteria aggregation model is first constructed and then applied

on set A to get information about the comprehensive preference

Disaggregation-aggregation (or regression) paradigm:

The comprehensive preference on a subset ARA is known a priori and

a consistent criteria aggregation model is inferred from this information

Page 8: Roman Słowiński Poznań University of Technology, Poland

8

Criteria aggregation model = preference model

The disaggregation-aggregation paradigam has been introduced

to MCDA by Jacquet-Lagreze & Siskos (1982) in the UTA method

– the inferred criteria aggregation

model is the additive value function with piecewise-linear marginal

value functions

The disaggregation-aggregation paradigam is consistent with the

„posterior rationality” principle by March (1988) and the inductive

learning used in artificial intelligence and knowledge discovery

Page 9: Roman Słowiński Poznań University of Technology, Poland

9

The comprehensive preference information is given in form of

a complete preorder on a subset of reference alternatives

ARA,

AR={a1,a2,...,am} – the reference alternatives are rearranged such

that ak ak+1 , k=1,...,m-1

Principle of the UTA method (Jacquet-Lagreze & Siskos, 1982)

A

AR

a1

a2

a5

a6a7

a3a4

Page 10: Roman Słowiński Poznań University of Technology, Poland

10

Example:

Let AR={a1, a2, a3}, G={Gain_1, Gain_2}

Evaluation of reference alternatives on criteria Gain_1, Gain_2:

Reference ranking:

Principle of the UTA method

Gain_1 Gain_2

a1 4 6

a2 5 5

a3 6 4

a1

a2

a3

Page 11: Roman Słowiński Poznań University of Technology, Poland

11

Principle of the UTA method

Page 12: Roman Słowiński Poznań University of Technology, Poland

12

Principle of the UTA method

Page 13: Roman Słowiński Poznań University of Technology, Poland

13

Let’s change the reference ranking:

One linear piece per each marginal value function u1, u2 is not enough

Principle of the UTA method

Gain_1 Gain_2

a1 4 6

a2 5 5

a3 6 4

a1

a3

a2

a1

a2

a3

u1=w1Gain_1, u2=w2Gain_2, U=u1+u2

For a1a3, w2>w1,

but for a3a2, w1>w2,

thus, marginal value functions cannot be linear

Page 14: Roman Słowiński Poznań University of Technology, Poland

14

Principle of the UTA method

Page 15: Roman Słowiński Poznań University of Technology, Poland

15

Principle of the UTA method

The inferred value of each reference alternative aAR:

where

is a calculated value function,

is a value function compatible with

the

reference ranking,

+ and - are potential errors of over- and under-estimation of the

compatible value function, respectively.

The intervals [i, i] are divided into (i–1) equal sub-intervals with

the end points (i=1,...,n)

n

iii ag'ua'U

1

g

aaaUa'U gg

n

iii aguaU

1

g

iiii

iji ,...,j,

jg

11

1

Page 16: Roman Słowiński Poznań University of Technology, Poland

16

Principle of the UTA method

The marginal value of alternative aA is approximated by a linear

interpolation: for 1 ji

jii g,gag

jii

jiij

iji

jiij

iiii gugugg

gagguagu

11

Page 17: Roman Słowiński Poznań University of Technology, Poland

17

Principle of the UTA method

Ordinal regression principle

if then one of the following holds

N.B. In practice, „0” is replaced here by a small positive number that may

influence the result

Monotonicity of preferences

Normalization

11

11

0

0

k~kkk

kkkk

aaa,a

aaa,a

iff

iff

11 kkkk a'Ua'Ua,a gg

n,...,i;,...,jgugu ijii

jii 11101

n,...,iu

u

ii

n

iii

10

11

Page 18: Roman Słowiński Poznań University of Technology, Poland

18

Principle of the UTA method

The marginal value functions (breakpoint variables) are estimated by

solving the LP problem

11

11

0

0

k~kkk

kkkk

Aa

aaa,a

aaa,a

aaFR

iff

iff

to subject

Min

n,...,i;,...,jgugu ijii

jii 11101

ji,Aa,a,a,gu

n,...,iu

u

Rjii

ii

n

iii

and

000

10

11

(C)

Page 19: Roman Słowiński Poznań University of Technology, Poland

19

Principle of the UTA method

If F*=0, then the polyhedron of feasible solutions for ui(gi) is not empty and

there exists at least one value function U[g(a)] compatible with the

complete preorder on AR

If F*>0, then there is no value function U[g(a)] compatible with the

complete preorder on AR – three possible moves:

increasing the number of linear pieces i for ui(gi)

revision of the complete preorder on AR

post optimal search for the best function with respect to Kendall’s in the area F F*+

Jacquet-Lagreze & Siskos (1982)

F F*+

polyhedron of constraints (C)

F= F*

Page 20: Roman Słowiński Poznań University of Technology, Poland

20

Współczynnik Kendalla

Do wyznaczania odległości między preporządkami stosuje się miarę Kendalla

Przyjmijmy, że mamy dwie macierze kwadratowe R i R* o rozmiarze m m, gdzie m = |AR|, czyli m jest liczbą wariantów referencyjnych

macierz R jest związana z porządkiem referencyjnym podanym przez decydenta,

macierz R* jest związana z porządkiem dokonanym przez funkcję użyteczności wyznaczoną z zadania PL (zadania regresji porządkowej)

Każdy element macierzy R, czyli rij (i, j=1,..,m), może przyjmować wartości:

To samo dotyczy elementów macierzy R*

Tak więc w każdej z tych macierzy kodujemy pozycję (w porządku) wariantu a względem wariantu b

gdy ,1

gdy ,50

gdy ,0

ji

i~j

ij

ij

aa

aa.

aalubji

r

Page 21: Roman Słowiński Poznań University of Technology, Poland

21

Współczynnik Kendalla

Następnie oblicza się współczynnik Kendalla :

gdzie dk(R,R*) jest odległością Kendalla między macierzami R i R*:

Stąd -1, 1

Jeżeli = -1, to oznacza to, że porządki zakodowane w macierzach R i R*

są zupełnie odwrotne, np. macierz R koduje porządek a b c d,

a macierz R* porządek d c b a

Jeżeli = 1, to zachodzi całkowita zgodność porządków z obydwu macierzy.

W tej sytuacji błąd estymacji funkcji użyteczności F*=0

W praktyce funkcję użyteczności akceptuje się, gdy 0.75

1

41

mm

*R,Rdk

m

i

m

jijijk *rr*R,Rd

1 121

Page 22: Roman Słowiński Poznań University of Technology, Poland

22

Example of UTA+

Ranking of 6 means of transportation

1

2

Page 23: Roman Słowiński Poznań University of Technology, Poland

23

Preference attitude: „economical”

Page 24: Roman Słowiński Poznań University of Technology, Poland

24

Preference attitude: „hurry”

Page 25: Roman Słowiński Poznań University of Technology, Poland

25

Preference attitude: „hurry”