28
Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology, Poland Salvatore Greco Faculty of Economy, University of Catania, Italy Roman Słowiński Institute of Computing Science, Poznań University of Technology, Poland

Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

Embed Size (px)

Citation preview

Page 1: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

Dominance-Bases Rough Set Approach: Features, Extensions and Application

Krzysztof Dembczyński Institute of Computing Science,Poznań University of Technology, Poland

Salvatore GrecoFaculty of Economy, University of Catania, Italy

Roman SłowińskiInstitute of Computing Science,Poznań University of Technology, Poland

Page 2: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

2

Topics

Philosophy of Dominance-Based Rough Set Approach (DRSA)

Preliminaries of DRSA

Extensions of DRSA

Variable Consistency DRSA

Multi-Valued DRSA

Continuous Decision Criterion and DRSA

Conclusion

Page 3: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

3

The Philosophy of Dominance-Based Rough Set Approach

The aim of the decision analysis is to answer two questions:

To explain decisions in terms of the circumstances in which they were made.

To give a recommendation how to make a good decision under specific circumstances.

One of decision problems is the multicriteria sorting

Multicriteria sorting concerns an assignment of the objects to pre-defined classes (concepts) that are preference-ordered.

Page 4: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

4

The Philosophy of Dominance-Based Rough Set Approach

Analyzed objects are described using criteria

Criteria are attributes with preference-ordered domain

Decision criterion shows the class of any object

Multicriteria decision problem has no solution unless a preference model is defined

Functional

Relational

Decision rules

Page 5: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

5

The Philosophy of Dominance-Based Rough Set Approach

Data are very often inconsistent with dominance principle that requires that an object having a better (not worse) evaluation on considered criteria cannot be assigned to a worse class.

H I G H

L O W

Page 6: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

6

The Philosophy of Dominance-Based Rough Set Approach

Greco, Matarazzo and Słowiński have proposed Dominance-Based Rough Set Approach

The Classical Rough Set Approach, proposed by Pawlak, has been proved as excellent tool for data analysis, however, it was falling for multicriteria sorting problem

The analyzed objects may be considered only in the perspective of available information

Page 7: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

7

The Philosophy of Dominance-Based Rough Set Approach

The rough set approaches features:

Information has granular structure

Approximation of one knowledge by another knowledge

Analysis of uncertain and inconsistent data

Inducing of “if…, then” decision rules

In DRSA the set of decision rules plays a role of comprehensive preference model

The rules syntax is concordant with Dominance Principle

Page 8: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

8

Topics

Philosophy of Dominance-Based Rough Set Approach (DRSA)

Preliminaries of DRSA

Extensions of DRSA

Variable Consistency DRSA

Multi-Valued DRSA

Continuous Decision Criterion and DRSA

Conclusion

Page 9: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

9

Preliminaries of DRSA

Basic notions

Outranking relation

x is at least so good as y with respect to criterion q

Dominance relation (reflexive and transitive)

x dominates y when on all criteria x outranks y (x is at least so good then y)

Data are often presented as a table

Because of preference order of classes it is possible to consider upward and downward unions of classes

Page 10: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

10

Preliminaries of DRSA

An ExampleFirst Criterion Second Criterion Decision Criterion

34.4 23.4 High

30.3 22.1 High

25 19 High

20 17 High

22 19.5 Medium

12 25 Medium

15.5 16.8 Medium

17.1 17.6 Medium

8.9 10.1 Low

11 13.5 Low

9 7 Low

12.5 4 Low

Page 11: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

11

Preliminaries of DRSA

An Example

c1

c20 40

40

20

20

+

+

++

o

oo

o

--- -

+

+

+ +

+

+

+ +

+

+

BEST

WORST

o

oo

oo

o

--- -

o

o

Page 12: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

12

Preliminaries of DRSA

Granules of Knowledge: Dominating and Dominated Sets

c1

c20 40

40

20

20

+

+

++

o

oo

o

--- -

+

+

+ +

+

+

+ +

+

+

BEST

WORST

o

oo

oo

o

--- -

o

o

Page 13: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

13

Preliminaries of DRSA

Granules of Knowledge: Dominating and Dominated Sets

c1

c20 40

40

20

20

+

+

++

o

oo

o

--- -

+

+

+ +

+

+

+ +

+

+

BEST

WORST

o

oo

oo

o

--- -

o

o

Page 14: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

14

Preliminaries of DRSA

Lower and Upper Approximation of the class unions

c1

c20 40

40

20

20

+

+

++

o

oo

o

--- -

+

+

+ +

+

+

+ +

+

+

BEST

WORST

o

oo

oo

o

--- -

o

o

Page 15: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

15

Preliminaries of DRSA

c1

c20 40

40

20

20

+

+

++

o

oo

o

--- -

+

+

+ +

+

+

+ +

+

+

WORST

o

oo

oo

o

--- -

o

o

BEST

Inducing of Decision Rules

Page 16: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

16

Preliminaries of DRSA

Form of Decision Rules

if f(x, c1) 25 and f(x, c2) 19, then x is at least High

if f(x, c1) 20 and f(x, c2) 17, then x could be at least High

if f(x, c1) 20 and f(x, c2) 17 and f(x, c1) 22 and f(x, c2) 19.5,

then x belongs to High or Medium

Page 17: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

17

Preliminaries of DRSA

c1

c20 40

40

20

20

+

+

++

o

oo

o

--- -

+

+

+ +

+

+

+ +

+

+

WORST

o

oo

oo

o

--- -

o

o

BEST

Inducing of Decision Rules with Hyperplanes

Page 18: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

18

Preliminaries of DRSA

Inducing of Decision Rules with Hyperplanes

c1

c20 40

40

20

20

+

+

++

o

oo

o

--- -

+

+

+ +

+

+

+ +

+

+

WORST

o

oo

oo

o

--- -

o

o

BEST

Page 19: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

19

Preliminaries of DRSA

Features

Analysis of multicriteria sorting problems with inconsistent information

It is possible to analyze objects described by criteria and regular attributes

Continuous domain of criteria (discretization is not needed)

Sorting of new objects

Page 20: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

20

Topics

Philosophy of Dominance-Based Rough Set Approach (DRSA)

Preliminaries of DRSA

Extensions of DRSA

Variable Consistency DRSA

Multi-Valued DRSA

Continuous Decision Criterion and DRSA

Conclusion

Page 21: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

21

Variable-Consistency DRSA

c1

c20 40

40

20

20

+

+

++

o

oo

o

--- -

+

+

+ +

+

+

+ +

+

+

BEST

WORST

o

oo

oo

o

--- -

o

o

Lower Approximation consists of limited counterexamples controlled by pre-defined level of certainty

Page 22: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

22

Multi-Valued DRSA

C1 C2 Decision

34.4 23.4 High

30.3 22.1 High

25-21 19 High

20-17 17 High

18-15 19.5 Medium

12 25 Medium

15.5 16.8 Medium

17.1 17.6 Medium

8.9 10.1 Low

11 13.5 Low

9 7 Low

12.5 4 Low

Interval order

object x is not worse than

y with respect to a single criterion, if there exist a value describing x that is not worse than at least one value describing y

Form of the rules:

if u(x) 21 then,

x is at least High

Page 23: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

23

Extensions of DRSA

VC-DRSA and MV-DRSA are only examples of extensions of DRSA.

Another example is the methodology that allows deal with missing values

There exist different strategies of induction of decision rules

It is also possible to induces decision trees using rough approximations

Page 24: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

24

Topics

Philosophy of Dominance-Based Rough Set Approach (DRSA)

Preliminaries of DRSA

Extensions of DRSA

Variable Consistency DRSA

Multi-Valued DRSA

Continuous Decision Criterion and DRSA

Conclusion

Page 25: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

25

Continuous Decision Criterion

C1 C2 D1 DC

34.4 23.4 High 34.5

30.3 22.1 High 31.5

25 19 High 25.4

20 17 High 22.1

22 19.5 Medium 21.5

12 25 Medium 20.1

15.5 16.8 Medium 17.4

17.1 17.6 Medium 16

8.9 10.1 Low 10.7

11 13.5 Low 9.5

9 7 Low 4.3

12.5 4 Low 3.5

What we can do?

Pre-discretization of decision criterion

Or

Analyzing data with continues decision

Large number of classes and unions of classes?

This is more inconsistencies

Looking for good association on the conditional part of the decision table

Page 26: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

26

Continuous Decision Criterion

Decision Rules

if f(x, c1) 34.4, then x is at least 34.5

if f(x, c2) 25, then x is at least 25.4

if f(x, c1) 20, then x is at least 21.5

if f(x, c1) 8.9, then x is at least 4.3

if f(x, c1) 17.1, then x is at most 20.1

Page 27: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

27

Topics

Philosophy of Dominance-Based Rough Set Approach (DRSA)

Preliminaries of DRSA

Extensions of DRSA

Variable Consistency DRSA

Multi-Valued DRSA

Discussion about Continuous Decision Criterion and DRSA

Conclusion

Page 28: Dominance-Bases Rough Set Approach: Features, Extensions and Application Krzysztof Dembczyński Institute of Computing Science, Poznań University of Technology,

28

Conclusion

It is proven that:

The preference model in the form of rules derived from examples is more general then the classic functional or relational model and it is more understandable for the users because of its natural syntax.

It fulfils both explanation and recommendation tasks that are principal aims of decision analysis.

DRSA is still developing

DRSA in the Malaria Vulnerability Case Study in IIASA during YSSP