overview of rough sets
DESCRIPTION
Overview of Rough Sets. Rough Sets (Theoretical Aspects of Reasoning about Data) by Zdzislaw Pawlak. Contents. Introduction Basic concepts of Rough Sets information system equivalence relation / equivalence class / indiscernibility relation - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/1.jpg)
Rough
Set
Overview of Rough Sets
Rough Sets (Theoretical Aspects of Reasoning about Data)by Zdzislaw Pawlak
![Page 2: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/2.jpg)
2
Contents
1. Introduction 2. Basic concepts of Rough Sets
information system equivalence relation / equivalence class / indiscernibility relation set approximation (Lower & Upper Approximations)
- accuracy of Approximation- extension of the definition of approximation of sets
dispensable & indispensable- reducts and core
dispensable & indispensable attributes- independent- relative reduct & relative core
dependency in knowledge- partial dependency of attribute (knowledge)- significance of attributes
discernibility Matrix
3. Example decision Table dissimilarity Analysis
![Page 3: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/3.jpg)
3
Introduction
Rough Set theory
by Zdzislaw Pawlak in the early 1980’s
use : AI, information processing, data mining, KDD etc.
ex) feature selection, feature extraction, data reduction,
decision rule generation and pattern extraction (association rules) etc.
theory for dealing with information with uncertainties.
reasoning from imprecision data.
more specifically, discovering relationships in data.
The idea of the rough set consists of the approximation of a set(X) by a p
air of sets, called the low and the upper approximation of this set(X)
![Page 4: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/4.jpg)
4
Information System
Knowledge Representation System ( KR-system , KRS )
Information systems I = < U, Ω>
a finite set U of objects , U={x1, x2, .., xn} ( universe )
a finite set Ω of attributes , Ω = {q1, q2, .., qm} ={ C, d }
C : set of condition attribute d : decision attribute
ex) I = < U, {a, c, d}>
UU a c dx1 1 4 yes
x2 1 1 no
x3 2 2 no
x4 2 2 Yes
x5 3 3 no
x6 1 3 yes
x7 3 3 no
![Page 5: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/5.jpg)
5
Equivalence Relation
An equivalence relation R on a set U is defined as
i.e. a collection R of ordered pairs of elements of U, satisfying certain properties.
1. Reflexive: xRx for all x in U ,
2. Symmetric: xRy implies yRx for all x, y in U
3. Transitive: xRy and yRz imply xRz for all x, y, z in U
))}()((|),{(R ycxcUUyx jjc j
![Page 6: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/6.jpg)
6
Equivalence Class
R : any subset of attributes( )
If R is an equivalence relation over U,
then U/R is the family of all equivalence classes of R
: equivalence class in R containing an element
ex) an subset of attribute ‘R1={a}’ is equivalence relation
the family of all equivalence classes of {a}
: U/ R1 ={{ 1, 2, 6}{3, 4}{5, 7}}
equivalence class
:
※ A family of equivalence relation over U will be call a knowledge base over U
UU a c d1 1 4 yes
2 1 1 no
3 2 2 no
4 2 2 Yes
5 3 3 no
6 1 3 yes
7 3 3 no
R
Rx][ Ux
}6,2,1{]1[1R }4,3{]3[
1R }7,5{]5[
1R
![Page 7: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/7.jpg)
7
Indiscernibility relation
If and ,
then is also an equivalence relation, ( IND(R) )
and will be called an indiscernibility relation over R ※ : intersection of all equivalence relations belonging to R
equivalence class of the equivalence relation IND(R)
:
U/IND(R) : the family of all equivalence classes of IND(R)
ex) an subset of attribute ‘R={a, c}’
the family of all equivalence classes of {a}
: U/{a}={{ 1, 2, 6}{3, 4}{5, 7}}
the family of all equivalence classes of {b}
: U/{c}={{ 2}{3, 4}{5, 6, 7}{1}}
the family of all equivalence classes of IND(R)
: U/IND(R) ={{2}{6}{1} {3,4}{5,7}}=U/{a,c}
UU a c d
1 1 4 yes
2 1 1 no
3 2 2 no
4 2 2 Yes
5 3 3 no
6 1 3 yes
7 3 3 no
R)R( ][][ xx IND
R
R R
R
![Page 8: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/8.jpg)
8
I = <U, Ω> = <U, {a, c}>
U is a set (called the universe) Ω is an equivalence relation on U (called an indiscernibility relation).
U is partitioned by Ω into equivalence classes,
elements within an equivalence class are indistinguishable in I.
An equivalence relation induces a partitioning of the universe.
The partitions can be used to build new subsets of the universe.
※ equivalence classes of IND(R) are called basic categories (concepts) of knowledge R
※ even union of R-basic categories will be called R-category
UU a c1 1 4
2 1 1
3 2 2
4 2 2
6 1 3
5 3 3
7 3 3
![Page 9: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/9.jpg)
9
Set Approximation
Given I = <U, Ω> Let and R : equivalence relation We can approximate X using only the information contained in R
by constructing the R-lower( ) and R-upper( ) approximations of X,
where
or
X is R-definable (or crisp) if and only if ( i.e X is the union of some R-basic categories, called R-definable set, R-exact set)
X is R-undefinable (rough) with respect to R if and only if ( called R-inexact, R-rough)
UX
XR XR
}.][|{ XxxXR R
},][|{ XxxXR R
}.:/{ XYRUYXR
}:/{ XYRUYXR
XRXR
XRXR
)I(INDR
![Page 10: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/10.jpg)
10
R-positive region of X : R-borderline region of X : R-negative region of X :
XRXRXBNR )(
XRUXNEGR )(
XRXPOSR )(
U
U/R R : subset of attributes
set X
XRXR ∴ X is R-definable
U/R
U
set X
∴ X is R-rough (undefinable)XR XR
![Page 11: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/11.jpg)
11
EX) I = <U, Ω>, let R={a, c} , X={x | d(x) = yes}={1, 4, 6}
► approximate set X using only the information contained in R
the family of all equivalence classes of IND(R)
: U/IND(R) = U/R = {{1}{ 2}{6} {3,4}{5,7}}
R-lower approximations of X
:
R-lower approximations of X
:
※ The set X is R-rough since the boundary region is not empty
UU a c d
1 1 4 yes
2 1 1 no
3 2 2 no
4 2 2 Yes
5 3 3 no
6 1 3 yes
7 3 3 no}4,3,6,1{}][|{R XxxX B
}6,1{}][|{R XxxX B
}4,3{)(R XBN }7,5,2{)(R XNEG}6,1{)(R XPOS
![Page 12: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/12.jpg)
12
Lower & Upper Approximations
yes
yes/no
no
{x1, x6}
{x3, x4}
{x2, x5,x7}
XR
XR
![Page 13: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/13.jpg)
13
Accuracy of Approximation
accuracy measure αR(X)
: the degree of completeness of our knowledge R about the set X
If , the R-borderline region of X is empty
and the set X is R-definable (i.e X is crisp with respect to R).
If , the set X has some non-empty R-borderline region
and X is R-undefinable (i.e X is rough with respect to R).
ex) let R={a, c} , X={x | d(x) = yes}={1, 4, 6}
Rcard
RcardXR )( .X.10 R
1)( XR
1)( XR
}4,3,6,1{}][|{R XxxX B
}6,1{}][|{R XxxX B
5.04
2)(R
Rcard
RcardX
![Page 14: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/14.jpg)
14
R-roughness of X
: the degree of incompleteness of knowledge R about the set X
ex) let R={a, c} , X={x | d(x) = yes}={1, 4, 6}
Y={x | d(x) = no}={2, 3, 6, 7}
U/IND(R) = U/R = {{1}{ 2}{6} {3,4}{5,7}}
)(1)( XX RR
5.05.01)(1)(R XX R
}4,3,6,1{}][|{R R XxxX
}6,1{}][|{R R XxxX
5.04/2)(R X
}6,2{}][|{R R YxxY
}7,5,6,4,3,2{}][|{R R YxxY
67.06/4)(1)( RR YY 33.06/2)(R Y
![Page 15: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/15.jpg)
15
Extension of the definition of approximation of sets
F={X1, X2, ..., Xn} : a family of non-empty sets and
=> R-lower approximation of the family F : R-upper approximation of the family F :
},,{ 21 nXRXRXRFR
},,,{ 11 nXRXRXRFR
ex) R={a, c}
F={X, Y}={{1,4,6}{2,3,5,7}} , X={x | d(x) = yes},
Y={x | d(x) = no}
U/IND(R) = U/R = {{1}{ 2}{6} {3,4}{5,7}}
}7,6,5,2,1{}}7,5,2}{6,1{{},{ YRXRFR
}7,6,5,4,3,2,1{}}7,6,5,4,3,2}{6.4.3.1{{},{ YRXRFR
UF
UU a c d
1 1 4 yes
2 1 1 no
3 2 2 no
4 2 2 Yes
5 3 3 no
6 1 3 yes
7 3 3 no
![Page 16: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/16.jpg)
16
the accuracy of approximation of F: the percentage of possible correct decisions when classifying objects employing the knowledge R
the quality of approximation of F : the percentage of objects which can be correctly classified to classes of F
employing the knowledge R
i
iR
XRcard
XRcardF )(
Ucard
XRcardF i
R)(
7/5)( FR2/1)64/()32()( FR
ex) R={a, c}
F={X, Y}={{1,4,6}{2,3,6,7}} , X={x | d(x) = yes}, Y={x | d(x) = no}
}7,6,5,2,1{}}7,5,2}{6,1{{},{ YRXRFR
}7,6,5,4,3,2,1{}}7,6,5,4,3,2}{6.4.3.1{{},{ YRXRFR
![Page 17: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/17.jpg)
17
Dispensable & Indispensable
Let R be a family of equivalence relations let if IND(R) = IND(R-{a}), then a is dispensable in R if IND(R) ≠ IND(R-{a}), then a is indispensable in R
the family R is independent if each is indispensable in R ; otherwise R is dependent
ex) R={a, c}
U/{a}={{ 1, 2, 6}{3, 4}{5, 7}}
U/{b}={{ 2}{3, 4}{5, 6, 7}{1}}
U/IND(R) ={{1}{ 2}{6} {3,4}{5,7}}=U/{a,b}
∴ a, b : indispensable in R
∴ R is independent (∵ U/IR ≠ U/{b}, U/IR ≠ U/{a})
Ra
Ra
UU a c1 1 4
2 1 1
3 2 2
4 2 2
5 3 3
6 1 3
7 3 3
![Page 18: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/18.jpg)
18
Core & Reduct
the set of all indispensable relation in R => the core of R , ( CORE(R) ) is a reduct of R if Q is independent and IND(Q) = IND(R) , ( RED(R) )RQ
)R()R( REDCORE
ex) a family of equivalence relations R={P, Q, R}
U/P ={{1,4,5}{2,8}{3}{6,7}}
U/Q ={{1,3,5}{6}{2,4,7,8}}
U/R ={{1,5}{6}{2,7,8}{3,4}}
U/{P,Q}={{1,5}{4}{{2,8}{3}{6}{7}}
U/{P,R}={{1,5}{4}{2,8}{3}{6}{7}}
U/{Q,R}={{1,5}{3}{6}{2,7,8}{4}}
U/R={{1,5}{6}{2,8}{3}{4}{7}}
U/{P,Q}= U/R =>R is dispensable in R
U/{P,R} }= U/R => Q is dispensable in R
U/{Q,R} }≠ U/R => P is indispensable in R
∴DORE(R) ={P}
∴RED(R) = {P,Q} and {P,R}
( U/{∵ P,Q}≠U/{P} , U/{P,Q}≠U/{Q}
U/{P,R}≠U/{P} , U/{P,R}≠U/{R} )
※ a reduct of knowledge is its essential part.
※ a core is in a certain sense its most important part.
![Page 19: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/19.jpg)
19
Dispensable & Indispensable AttributesLet R and D be families of equivalence relation over U,
if , then the attribute a is dispensable in I ,
if , then the attribute a is indispensable in I ,
The R-positive region of D :
.Ra
XPOSUX
D/R R)D(
)D()D( }){R(R aPOSPOS )D()D( }){R(R aPOSPOS
![Page 20: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/20.jpg)
20
ex) R={a, c} D={d}
U/{a}={{ 1, 2, 6}{3, 4}{5, 7}}
U/{c}={{ 2}{3, 4}{5, 6, 7}{1}}
U/D={{1,4,6}{2,3,5,7}}
U/IND(R) ={{1}{ 2}{6} {3,4}{5,7}}=U/{a,c}
UU a c d
1 1 4 yes
2 1 1 no
3 2 2 no
4 2 2 Yes
5 3 3 no
6 1 3 yes
7 3 3 no
}7,6,5,2,1{}}7,5,2}{6,,1{{R)D(D/
R
XPOSUX
}2,1{}}2}{1{{)D(}){R( aPOS
}7,5{}}7,5{{)D(}){R( cPOS
)D()D( }){R(R aPOSPOS
)D()D( }){R(R cPOSPOS
=> the relation ‘a’ is indispensable in R (‘a’ is indispensable attribute)
=> the relation ‘c’ is indispensable in R (‘c’ is indispensable attribute)
![Page 21: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/21.jpg)
21
Independent
If every c in R is D-indispensable, then we say that R is D-independent
(or R is independent with respect to D)
ex) R={a, c} D={d}
∴ R is D-independent ( , ))D()D( }){R(R aPOSPOS )D()D( }){R(R cPOSPOS
UU a c d
1 1 4 yes
2 1 1 no
3 2 2 no
4 2 2 Yes
5 3 3 no
6 1 3 yes
7 3 3 no
![Page 22: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/22.jpg)
22
Relative Reduct & Relative Core
The set of all D-indispensable elementary relation in R will be called the D-core of R, and will be denoted as CORED(R)
※ a core is in a certain sense its most important part.
The set of attributes is called a reduct of R,
if C is the D-independent subfamily of R and
=> C is a reduct of R ( REDD(R) )
※ a reduct of knowledge is its essential part.
※ REDD(R) is the family of all D-reducts of R
ex) R={a, c} D={d}
CORED(R) ={a, c} REDD(R) ={a, c}
RC
).D()D( RPOSPOSC
)()( RREDRCORED
![Page 23: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/23.jpg)
23
An Example of Reducts & Core
U Headache Muscle pain
Temp. Flu
U1 Yes Yes Normal No U2 Yes Yes High Yes U3 Yes Yes Very-high Yes U4 No Yes Normal No U5 No No High No U6 No Yes Very-high Yes
U={U1, U2, U3, U4, U5, U6} =let {1,2,3,4,5,6}
Ω={headache, Muscle pan, Temp, Flu}={a, b, c, d}
condition R={a, b, c}, decision D={d}
U/{a}={{1,2,3}{4,5,6}}
U/{b}={{1,2,3,4,6}{5}}
U/{c}={{1,4}{2,5}{3,6}}
U/{a,b}={1,2,3}{4,6}{5}}
U/{a,c}={{1}{2}{3}{4}{5}{6}}
U/{b,c}={{1,4}{2}{3,6}{5}}
U/R={{1}{4}{2}{5}{3}{6}}
U/D={{1,4,5}{2,3,6}}
POSR(D)={{1,4,5}{2,3,6}}={1,2,3,4,5,6}
POSR-{a}(D)={{1,4,5}{2,3,6}}={1,2,3,4,5,6}
POSR-{b}(D)={{{1,4,5}{2,3,6}}={1,2,3,4,5,6}
POSR-{c}(D)={{5}}={5}
• relation ‘a’, ‘b’ is dispensable• relation ‘c’ is indispensable
=> D-core of R =CORED(R)={c}
to find reducts of R={a, b, c}
• {a, c} is D-independent and POS{a, c}(D)=POSR(D)
(∵POS{a}(D)={} ≠POS{a, c}(D)
POS{c}(D)={1,4,3,6} ≠POS{a, c}(D) )
• {b, c} is D-independent and POS{b, c}(D)=POSR(D) => {a, c} {b, c} is the D-reduct of R
POSR-{ab}(D)={{1,4}{3,6}}={1,4,3,6}POSR-{ac}(D)={{5}}={5}POSR-{bc}(D)={}
![Page 24: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/24.jpg)
24
U Headache Muscle pain
Temp. Flu
U1 Yes Yes Normal No U2 Yes Yes High Yes U3 Yes Yes Very-high Yes U4 No Yes Normal No U5 No No High No U6 No Yes Very-high Yes
U Musclepain
Temp. Flu
U1,U4 Yes Normal No
U2 Yes High YesU3,U6 Yes Very-high YesU5 No High No
U Headache Temp. Flu
U1 Yes Norlmal NoU2 Yes High YesU3 Yes Very-high YesU4 No Normal NoU5 No High NoU6 No Very-high Yes
Reduct1 = {Muscle-pain,Temp.}
Reduct2 = {Headache, Temp.}CORE = {Headache, Temp} ∩ {Muscle Pain, Temp}
= {Temp}
![Page 25: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/25.jpg)
25
Dependency in knowledge
Given knowledge P, Q U/P={{1,5}{2,8}{3}{4}{6}{7}} U/Q={{1,5}{2,7,8}{3,4,6}}
If , then Q depends on P (P Q)⇒)Q()P( INDIND
![Page 26: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/26.jpg)
26
Partial Dependency of knowledge
I=<U, Ω> and Knowledge Q depends in a degree k (0≤k≤1 ) from knowledge P
(P⇒k Q)
ex) U/Q={{1}{2,7}{3,6}{4}{5,8}}
U/P={{1,5}{2,8}{3}{4}{6}{7}}
POSP(Q) = {3,4,6,7}
the degree of dependency between Q and P
: (P⇒0.5 Q )
If k = 1 we say that Q depends totally on P. If k < 1 we say that Q depends partially (in a degree k) on P.
QP,
Ucard
POScardk
)Q()Q( P
P
5.08
4)Q()Q( P
P Ucard
POSk
![Page 27: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/27.jpg)
27
Significance of attributes
ex) R={a, b, c}, decision D={d}
U/{a}={{1,2,3}{4,5,6}}
U/{b}={{1,2,3,4,6}{5}}
U/{c}={{1,4}{2,5}{3,6}}
U/{a,b}={1,2,3}{4,6}{5}}
U/{a,c}={{1}{2}{3}{4}{5}{6}}
U/{b,c}={{1,4}{2}{3,6}{5}}
U/R={{1}{4}{2}{5}{3}{6}}
U/D={{1,4,5}{2,3,6}}
POSR(D)={{1,4,5}{2,3,6}}={1,2,3,4,5,6}
POSR-{a}(D)={{1,4,5}{2,3,6}}={1,2,3,4,5,6}
POSR-{b}(D)={{{1,4,5}{2,3,6}}={1,2,3,4,5,6}
POSR-{c}(D)={{5}}={5}
011)()( {b}-RR DD significance of attribute ‘b’ :
83.06/56/11)()( {c}-RR DD significance of attribute ‘c’ :
∴ the attribute c is most significant, since it most changes the positive region of U/IND(D)
011)()( {a}-RR DD significance of attribute ‘a’ :
![Page 28: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/28.jpg)
28
Discernibility Matrix
Let I = (U, Ω) be a decision table,
with U={x1, x2, .., xn}
C={a, b, c} : condition attribute set , D={d} : decision attribute set
By a discernibility matrix of I, denoted M(I)={mij}n×n
mij is the set of all the condition attributes that classify objects xi and xj into different classes.
U1 U2 U3 U4 U5 U6U1U2 cU3 c -U4 - a, c a, c
U5 - a, b a, b, c -
U6 a, c - - c b, c
- : same equivalence classes of the relation IND(d)
< Decision Table >
U Headache Muscle pain
Temp. Flu
U1 Yes Yes Normal No U2 Yes Yes High Yes U3 Yes Yes Very-high Yes U4 No Yes Normal No U5 No No High No U6 No Yes Very-high Yes
(a) (b) (d)(c)< Discernibility Matrix >
![Page 29: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/29.jpg)
29
Compute value cores and value reducts from the M(I)
the core can be defined now as the set of all single element entries of the discernibility matrix,
is the reduct of R, if B is the minimal subset of R such that
for any nonempty entry c ( ) in M(I)
},),(:{)( jisomeforamRaRCORE ij
RB
cB c
U1 U2 U3 U4 U5
U2 c
U3 c -
U4 - a, c a, c
U5 - a, b a, b, c -
U6 a, c - - c b, c
d-CORE(R) d-reducts : {a, c} {b, c}
![Page 30: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/30.jpg)
CHAPTER 6. Decision Tables
![Page 31: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/31.jpg)
31
• Proposition 6.2Each decision table can be uniquely decomposed
into two decision tables and such that in and in , where and
– compute the dependency between condition and decision attributes
– decompose the table into two subtables
),,,( DCAUT ),,,( 11 DCAUT
),,,( 22 DCAUT DC 1 1T DC 0 2T)(1 DPOSU C
)(/2 )(
DINDUXC XBNU
![Page 32: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/32.jpg)
32
• Example 1.
U a b c d e
1 1 0 2 2 0
2 0 1 1 1 2
3 2 0 0 1 1
4 1 1 0 2 2
5 1 0 2 0 1
6 2 2 0 1 1
7 2 1 1 1 2
8 0 1 1 0 1
condition
attribute
decision attribute
U a b c d e
3 2 0 0 1 1
4 1 1 0 2 2
6 2 2 0 1 1
7 2 1 1 1 2
U a b c d e
1 1 0 2 2 0
2 0 1 1 1 2
5 1 0 2 0 1
8 0 1 1 0 1
Table 2
Table 3
Table 1
• Table 2 is consistent, Table 3 is totally inconsistent
→ All decision rules in Table 2 are consistent
All decision rules in Table 3 are inconsistent
![Page 33: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/33.jpg)
33
• simplification of decision tables : reduction of condition attributes• steps
1) Computation of reducts of condition attributes which is equivalent to elimination of some column from the decision tables
2) Elimination of duplicate rows
3) Elimination of superfluous values of attributes
![Page 34: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/34.jpg)
34
• Example 2
U a b c d e
1 1 0 0 1 1
2 1 0 0 0 1
3 0 0 0 0 0
4 1 1 0 1 0
5 1 1 0 2 2
6 2 1 0 2 2
7 2 2 2 2 2
U a b d e
1 1 0 1 1
2 1 0 0 1
3 0 0 0 0
4 1 1 1 0
5 1 1 2 2
6 2 1 2 2
7 2 2 2 2
condition
attribute
decision attribute
e-dispensable condition attribute is c.
let R={a, b, c, d}, D={e}
CORED(R) ={a, b, d}
REDD(R) ={a, b, d}
remove column c
![Page 35: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/35.jpg)
35
• we have to reduce superfluous values of condition attributes in every decision rules
→ compute the core values1. In the 1st decision rules
• the core of the family of sets
• the core value is
U a b d e
1 1 0 1 1
2 1 0 0 1
3 0 0 0 0
4 1 1 1 0
5 1 1 2 2
6 2 1 2 2
7 2 2 2 2
}}4,1{},3,2,1{},5,4,2,1{{}]1[,]1[,]1{[ dbaF
dbadba ]1[]1[]1[]1[ },,{ }1{}4,1{}3,2,1{}5,4,2,1{
}2,1{]1[,1)1(,0)1(,1)1( edba
}1{}4,1{}3,2,1{]1[]1[ db
}4,1{}4,1{}5,4,2,1{]1[]1[ da
}2,1{}3,2,1{}5,4,2,1{]1[]1[ ba
0)1( b
![Page 36: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/36.jpg)
36
2. In the 2nd decision rules• the core of the family of sets
• the core value is
3. In the 3rd decision rules• the core of the family of sets
• the core value is
U a b d e
1 1 0 1 1
2 1 0 0 1
3 0 0 0 0
4 1 1 1 0
5 1 1 2 2
6 2 1 2 2
7 2 2 2 2
}}3,2{},3,2,1{},5,4,2,1{{}]1[,]1[,]1{[ dbaF
}2{]1[]1[]1[]1[ },,{ dbadba
}2,1{]1[,0)1(,0)1(,1)1( edba
}2,1{]1[]1[},2{]1[]1[},3,2{]1[]1[ badadb
1)1( a
}}3,2{},3,2,1{},3{{}]0[,]0[,]0{[ dbaF
}3{]0[]0[]0[]0[ },,{ dbadba
}4,3{]0[,0)0(,0)0(,0)0( edba
}3{]0[]0[},3{]0[]0[},3,2{]0[]0[ badadb
0)0( a
![Page 37: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/37.jpg)
37
4. In the 4th decision rules• the core of the family of sets
• the core value :
5. In the 5th decision rules• the core of the family of sets
• the core value is
U a b d e
1 1 0 1 1
2 1 0 0 1
3 0 0 0 0
4 1 1 1 0
5 1 1 2 2
6 2 1 2 2
7 2 2 2 2
}}4,1{},6,5,4{},5,4,2,1{{}]0[,]0[,]0{[ dbaF
}4{]0[]0[]0[]0[ },,{ dbadba
}4,3{]1[,1)0(,1)0(,1)0( edba
}5,4{]0[]0[},4,1{]0[]0[},4{]0[]0[ badadb
1)0(,1)0( db
}}7,6,5{},6,5,4{},5,4,2,1{{}]2[,]2[,]2{[ dbaF
}5{]2[]2[]2[]2[ },,{ dbadba
}7,6,5{]2[,2)2(,1)2(,1)2( edba
}5,4{]2[]2[},5{]2[]2[},6,5{]2[]2[ badadb
2)2( d
![Page 38: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/38.jpg)
38
6. In the 6th decision rules• the core of the family of sets
• the core value : not exist
7. In the 7th decision rules• the core of the family of sets
• the core value : not exist
U a b d e
1 1 0 1 1
2 1 0 0 1
3 0 0 0 0
4 1 1 1 0
5 1 1 2 2
6 2 1 2 2
7 2 2 2 2
}}7,6,5{},6,5,4{},7,6{{}]2[,]2[,]2{[ dbaF
}6{]2[]2[]2[]2[ },,{ dbadba
}7,6,5{]2[,2)2(,1)2(,2)2( edba
}6{]2[]2[},7,6{]2[]2[},6,5{]2[]2[ badadb
}}7,6,5{},7{},7,6{{}]2[,]2[,]2{[ dbaF
}7{]2[]2[]2[]2[ },,{ dbadba
}7,6,5{]2[,2)2(,2)2(,2)2( edba
}7{]2[]2[},7,6{]2[]2[},7{]2[]2[ badadb
U a b d e
1 - 0 - 1
2 1 - - 1
3 0 - - 0
4 - 1 1 0
5 - - 2 2
6 - - - 2
7 - - - 2
![Page 39: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/39.jpg)
39
• to compute value reducts– let’s compute value reducts for the ~1. 1st decision rules of the decision table
– 2 value reducts1. 2.
– Intersection of reducts : → core value
}2,1{]1[}1{}4,1{}3,2,1{]1[]1[ edb
eda ]1[}4,1{}4,1{}5,4,2,1{]1[]1[
edebea ]1[}4,1{]1[,]1[}3,2,1{]1[,]1[}5,4,2,1{]1[
1)1(and0)1( db0)1(and1)1( ba
0)1( b
eba ]1[}2,1{}3,2,1{}5,4,2,1{]1[]1[
![Page 40: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/40.jpg)
40
2. 2nd decision rules of the decision table
– 2 value reducts : – Intersection of reducts : → core value
3. 3rd decision rules of the decision table
– 1 value reduct : – Intersection of reducts : → core value
}2,1{]1[}3,2{]1[]1[ edb
ebaeda ]1[}2,1{]1[]1[,]1[}2{]1[]1[
edebea ]1[}3,2{]1[,]1[}3,2,1{]1[,]1[}5,4,2,1{]1[
0)1(and1)1(or0)1(and1)1( bada
0)1( a
}4,3{]0[}3,2{]0[]0[ edb
ebaeda ]0[}3{]0[]0[,]0[}3{]0[]0[
edebea ]0[}3,2{]0[,]0[}3,2,1{]0[,]0[}3{]0[
0)0( a0)0( a
![Page 41: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/41.jpg)
41
4. 4th decision rules of the decision table
– 1 value reduct : – Intersection of reducts : → core value
5. 5th decision rules of the decision table
– 1 value reduct :
– Intersection of reducts : → core value
}4,3{]0[}4{]0[]0[ edb
ebaeda ]0[}5,4{]0[]0[,]0[}4,1{]0[]0[
edebea ]0[}4,1{]0[,]0[}3,2,1{]0[,]0[}5,4,2,1{]0[
0)1(and1)1( db
2)2( d
0)1(and1)1( db
edebea ]2[}7,6,5{]2[,]2[}6,5,4{]2[},7,6,5{]0[}5,4,2,1{]2[
2)2( d
}7,6,5{]2[}5{]2[]2[},7,6,5{]2[}6,5{]2[]2[ 22 dadb
}7,6,5{]2[}5,4{]2[]2[ 2 ba
![Page 42: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/42.jpg)
42
6. 6th decision rules of the decision table
– 2 value reducts :
– Intersection of reducts : → core value : not exist
edebea ]2[}7,6,5{]2[,]2[}6,5,4{]2[},7,6,5{]2[}7,6{]2[
2)2(or2)2( da
}7,6,5{]2[}7,6{]2[]2[},7,6,5{]2[}6,5{]2[]2[ edaedb
}7,6,5{]2[}6{]2[]2[ eba
![Page 43: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/43.jpg)
43
7. 7th decision rules of the decision table
– 3 value reducts
– Intersection of reducts : → core value : not exist
– reducts : = 24 solutions to our problem
},7,6,5{]2[}7,6{]2[ ea
2)2(or2)2(or2)2( dba
edeb ]2[}7,6,5{]2[,]2[}7{]2[
U a b d e
1 1 0 Ⅹ 1
1′ Ⅹ 0 1 1
2 1 0 Ⅹ 1
2 ′ 1 Ⅹ 0 1
3 0 Ⅹ Ⅹ 0
4 Ⅹ 1 1 0
5 Ⅹ Ⅹ 2 2
6 Ⅹ Ⅹ 2 2
6 ′ 2 Ⅹ Ⅹ 2
7 Ⅹ Ⅹ 2 2
7 ′ Ⅹ 2 Ⅹ 2
7″ 2 Ⅹ Ⅹ 2
3211122
}7,6,5{]2[}7,6{]2[]2[},7,6,5{]2[}7{]2[]2[ edaedb
}7,6,5{]2[}7{]2[]2[ eba
![Page 44: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/44.jpg)
44
One solution Another solution
U a b d e
1 1 0 Ⅹ 1
2 1 Ⅹ 0 1
3 0 Ⅹ Ⅹ 0
4 Ⅹ 1 1 0
5 Ⅹ Ⅹ 2 2
6 Ⅹ Ⅹ 2 2
7 2 Ⅹ Ⅹ 2
U a b d e
1 1 0 Ⅹ 1
2 1 0 Ⅹ 1
3 0 Ⅹ Ⅹ 0
4 Ⅹ 1 1 0
5 Ⅹ Ⅹ 2 2
6 Ⅹ Ⅹ 2 2
7 Ⅹ Ⅹ 2 2
U a b d e
1,2 1 0 Ⅹ 1
3 0 Ⅹ Ⅹ 0
4 Ⅹ 1 1 0
5,6,7 Ⅹ Ⅹ 2 2
U a b d e
1 1 0 Ⅹ 1
2 0 Ⅹ Ⅹ 0
3 Ⅹ 1 1 0
4 Ⅹ Ⅹ 2 2
identical
enumeration is not essential
minimal solution
![Page 45: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/45.jpg)
45
10.4 Pattern Recognition[ Table10 ] : Digits display unit in a calculator
assumed to represent a characterization of “hand written” digits
UU a b c d e f G0 1 1 1 1 1 1 01 0 1 1 0 0 0 02 1 1 0 1 1 0 13 1 1 1 1 0 0 14 0 1 1 0 0 1 15 1 0 1 1 0 1 16 1 0 1 1 1 1 17 1 1 1 0 0 0 08 1 1 1 1 1 1 19 1 1 1 1 0 1 1
a
b
c
d
e
fg
▶ Out task is to find a minimal description of each digit
and corresponding decision algorithm.
![Page 46: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/46.jpg)
46
compute the core attributesUU b c d e f g0 1 1 1 1 1 01 1 1 0 0 0 02 1 0 1 1 0 13 1 1 1 0 0 14 1 1 0 0 1 15 0 1 1 0 1 16 0 1 1 1 1 17 1 1 0 0 0 08 1 1 1 1 1 19 1 1 1 0 1 1
UU a c d e f g0 1 1 1 1 1 01 0 1 0 0 0 02 1 0 1 1 0 13 1 1 1 0 0 14 0 1 0 0 1 15 1 1 1 0 1 16 1 1 1 1 1 17 1 1 0 0 0 08 1 1 1 1 1 19 1 1 1 0 1 1
[ drop attribute a ]
[ drop attribute b ]
decision rules are inconsistent
Rule1 : b1c1d0e0f0g0 → a0b1c1d0e0f0g0 Rule7 : b1c1d0e0f0g0 → a1b1c1d0e0f0g0
![Page 47: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/47.jpg)
47
UU a b d e f g0 1 1 1 1 1 01 0 1 0 0 0 0
2 1 1 1 1 0 13 1 1 1 0 0 14 0 1 0 0 1 15 1 0 1 0 1 16 1 0 1 1 1 17 1 1 0 0 0 08 1 1 1 1 1 19 1 1 1 0 1 1
UU a b c e f g0 1 1 1 1 1 01 0 1 1 0 0 02 1 1 0 1 0 13 1 1 1 0 0 14 0 1 1 0 1 15 1 0 1 0 1 16 1 0 1 1 1 17 1 1 1 0 0 08 1 1 1 1 1 19 1 1 1 0 1 1
←[drop attribute c]
[drop attribute d]→
UU a b c d f g0 1 1 1 1 1 01 0 1 1 0 0 02 1 1 0 1 0 13 1 1 1 1 0 14 0 1 1 0 1 15 1 0 1 1 1 16 1 0 1 1 1 17 1 1 1 0 0 08 1 1 1 1 1 19 1 1 1 1 1 1
UU a b c d e g0 1 1 1 1 1 01 0 1 1 0 0 02 1 1 0 1 1 13 1 1 1 1 0 14 0 1 1 0 0 15 1 0 1 1 0 16 1 0 1 1 1 17 1 1 1 0 0 08 1 1 1 1 1 19 1 1 1 1 0 1
[drop attribute f]→
←[drop attribute e]
decision rules are
consistent
decision rules are
inconsistent
![Page 48: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/48.jpg)
48
UU a b c d e f0 1 1 1 1 1 11 0 1 1 0 0 02 1 1 0 1 1 03 1 1 1 1 0 04 0 1 1 0 0 15 1 0 1 1 0 16 1 0 1 1 1 17 1 1 1 0 0 08 1 1 1 1 1 19 1 1 1 1 0 1
[drop attribute g]
∴ attribute c, d : dispensable
attribute a, b, e, f, g : indispensable
the set {a, b, e, f, g} : core
sole reducts : {a, b, e, f, g}
decision rules are inconsistent
![Page 49: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/49.jpg)
49
compute reduct • all attribute set : {a, b, c, d, e, f, g}• core : {a, b, e, f, g} • reduct : {a, b, e, f, g}
UU c d a b e f G0 1 1 1 1 1 1 01 1 0 0 1 0 0 02 0 1 1 1 1 0 13 1 1 1 1 0 0 14 1 0 0 1 0 1 15 1 1 1 0 0 1 16 1 1 1 0 1 1 17 1 0 1 1 0 0 08 1 1 1 1 1 1 19 1 1 1 1 0 1 1
![Page 50: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/50.jpg)
50
compute the core values of attributes for table11
UU a b e f G0 1 1 1 1 01 0 1 0 0 02 1 1 1 0 13 1 1 0 0 14 0 1 0 1 15 1 0 0 1 16 1 0 1 1 17 1 1 0 0 08 1 1 1 1 19 1 1 0 1 1
The core value in rule 1 and 4 : a0
The core value in rule 7 and 9 : a1
UU b a e f G0 1 1 1 1 01 1 0 0 0 02 1 1 1 0 13 1 1 0 0 14 1 0 0 1 15 0 1 0 1 16 0 1 1 1 17 1 1 0 0 08 1 1 1 1 19 1 1 0 1 1
The core value in rule 5 and 6 : b0
The core value in rule 8 and 9 : b1
[Table12 : Removing the attribute a ] [Table13 : Removing the attribute b ]
![Page 51: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/51.jpg)
51
UU e a b f G0 1 1 1 1 01 0 0 1 0 02 1 1 1 0 13 0 1 1 0 14 0 0 1 1 1
5 0 1 0 1 16 1 1 0 1 17 0 1 1 0 08 1 1 1 1 19 0 1 1 1 1
The core value in rule 2 ,6 and 8 : e0
The core value in rule 3 ,5 and 9 : e1
UU f a b e G0 1 1 1 1 01 0 0 1 0 02 0 1 1 1 13 0 1 1 0 14 1 0 1 0 15 1 1 0 0 16 1 1 0 1 17 0 1 1 0 08 1 1 1 1 19 1 1 1 0 1
[ Table14 : Removing the attribute e ][ Table 15 : Removing the attribute f ]
UU g a b e f0 0 1 1 1 11 0 0 1 0 02 1 1 1 1 03 1 1 1 0 04 1 0 1 0 15 1 1 0 0 16 1 1 0 1 17 0 1 1 0 08 1 1 1 1 19 1 1 1 0 1
The core value in rule 2 and 3 : f0The core value in rule 8 and 9 : f1
[ Table 16 : Removing the attribute g ]
![Page 52: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/52.jpg)
52
core value for all decision rules
UU a b e f G0 _ _ _ 0
1 0 _ _ _
2 _ 1 0 _
3 _ 0 0 1
4 0 _ _ _ _5 _ 0 0 _ _6 _ 0 1 _ _7 1 _ _ _ 08 _ 1 1 1 19 1 1 0 1 _
[Table17]: all core value for table11
• rule 2, 3, 5, 6, 8, 9 are consistent.
• rule 0, 1, 4, 7 are inconsistent.
• to make the rules consistent ⇒ adding proper additional attributes
UU a b e f G0 1 1 1 1 01 0 1 0 0 02 1 1 1 0 13 1 1 0 0 14 0 1 0 1 15 1 0 0 1 16 1 0 1 1 17 1 1 0 0 08 1 1 1 1 19 1 1 0 1 1
[Table11]
UU a b e f g0 x x 1 x 00’ x x x 1 01 0 x x 0 x1’ 0 x x x 02 x x 1 0 x3 x x 0 0 14 0 x x 1 X4’ 0 x x x 15 x 0 0 x x6 x 0 1 x x7 1 x 0 x 07’ 1 x x 0 08 x 1 1 1 19 1 1 0 1 x
[Table 18] All possible value reduct for table11
![Page 53: Overview of Rough Sets](https://reader031.vdocuments.us/reader031/viewer/2022013108/5681303b550346895d95d456/html5/thumbnails/53.jpg)
53
UU a b e f g0 x x 1 x 00’ x x x 1 01 0 x x 0 x1’ 0 x x x 02 x x 1 0 x3 x x 0 0 14 0 x x 1 X4’ 0 x x x 15 x 0 0 x x6 x 0 1 x x7 1 x 0 x 07’ 1 x x 0 08 x 1 1 1 19 1 1 0 1 x
Table18• We have 16=24 minimal decision algorithms.
• One of the possible reduced algorithms
: e1g0 (f1g0) → 0
a0f0 (a0g0) → 1
e1f0 → 2
e0f0g1 → 3
a0f1 (a0g1) → 4
b0e0 → 5
b0e1 → 6
a1e0g0 (a1f0g0) → 7
b1e1f1g1 → 8
a1b1e0f1 → 9
a
b
c
d
e
fg