2016 IEEE International Conference on Cloud Computing and Big Data Analysis

Definability of Rough Approximations for Binary Relations and Cloud Computing

Yu-Ru Syau Ly-Fie Sugianto

Department of Information Management Monash Business School
National Formosa University Monash University
Yunlin, Taiwan Victoria 3800, Australia

En-Bing Lin*
Department of Mathematics
Central Michigan University
Mt. Pleasant, Michigan, USA
e-mail: enbing.lin@cmich.edu

Abstract-We establish a characterization of upper considered by several authors in generalized rough set
approximation for binary relations. Based on this models (for example, see [7], [11]-[13]). Unlike the lower
characterization, the R-definability of Slowinski and and upper approximations in the classical rough set model,
Vanderpooten is alternatively defined in terms of lower and the generalized lower and upper approximations are not
upper approximations. In addition, some necessary conditions necessarily defmable.
for R-definability are presented. In this paper, we investigate the R-definability of
Slowinski and Vanderpooten [6]. A characterization of upper
Keywords-rough sets; lower and upper approximations;
approximation for binary relations is established. Based on
binary relations; deflnability
this characterization, the R-defmability of Slowinski and
I. INTRODUCTION Vanderpooten [6] is alternatively defined in terms of lower
and upper approximation. In addition, some necessary
Pawlak's rough set model [1], [2] intrinsically is a study conditions are given. Furthermore, one may apply rough
of an equivalence relation, namely, an approximation of approximation to deal with large scaled granular computing
equivalence classes. The rough set philosophy is founded on [14], which gives rise to an approach to work on cloud
the assumption that every object of the universe (of discourse) computing.
U is associated with some information (data, knowledge).
Objects characterized by the same information are II. PRELIMINARIES
indiscernible (similar) in view of available information about Let U be a fixed non-empty set, we refer it as the
them. Any set of all indiscernible (similar) objects is called universe (of discourse). The notation ZU denotes the power
an elementary set and considered as a basic granule of set of U. A binary relation on U is a subset R !:; UXU. The
knowledge about the universe [3]. inverse relation of R is the binary relation R-1 obtained by
A set X !:; U is definable if it is a finite union of reversing the ordered pairs of R.
elementary sets. If a set is not definable, it is called rough. Let R be a binary relation on U, and letx E U. The set
Roughness is basically meant to describe how rough a set is.
H is determined by a pair of sets, called the lower and upper Rs(x) = {y E UI(x,y) E R} (1)
approximations of the rough set. The lower approximation of
a set X is the greatest definable set contained in X, whereas is called the R-successor set, or the right R-neighborhood, of
the upper approximation of a set X is the least defmable set x. The set
containing X [4]. Accordingly, a set X!:; U is defmable iff Rp(x) = {y E UI(y,x) E R} (2)
(abbreviation for if and only it) its lower and upper
approximations are identical. is called the R-predecessor set, or the left R-neighborhood, of
Classical definitions of lower and upper approximations Note that Rs(x) Ri/(x) . By ER we mean the
x. =

[1], [2] were extended to the use of a general binary relation equivalence relation on U by regarding objects x, y of U as
[5 to 9] instead of an equivalence relation. Wang et al. [8] equivalent ifRs(x) Rs (y). The family VIER of all distinct
=

relaxed the equivalence relations to general binary relations, ER-equivalence classes is called the derived partition R [15].
and studied relation decision systems. For any X !:; U , the lower and upper approximations,
The Rough Set theoretical definability concept was R (X) and R*(X) , respectively, of X with respect to R­
*
generalized to tolerance relations (reflexive and symmetric successor sets are defined as follows [16], [17]:
relations) in [10], and to arbitrary binary relations, called R­
definability, in [6]. The notion of definability was also R (X)
*
= {x E UIRs(x) !:; X}, (3)

978-1-5090-2594-7116/$31.00 ©20 16 IEEE 233

 R*(X) = {x E UIRs(x) () x * 0}. (4)

Accordingly, we have R.(X) U -R*(U -X). In

U Rp(x) = UR* ({x}) = R*(X)
=
XEX XEX
addition, ={x E UIRs(x) () X * 0}.
R (0) {x E UIRs(x) 0} (5)
*
= =

IV. AN ALTERNATIVE DEFINITION OF R-DEFINABILITY

andR*(U) U -R (0) [18]. It is known [16, 17] that
*
=

Motivated by [13], we obtain, by Definition 1 and

R*(UXEdlX) UXEdlR*(X) 'if cA � 2u .
= (6) Theorem 2, an alternative defmition of R-definability as
follows.
Theorem 1 [6]. Considering a subsetX �U and a binary Definition 2. Considering a subset X � U and a binary
reflexive relation R defined on U, we have relation R on U, Xis R-definable if and only if:
UXEXRp(x) = {x E UIRs(x) () X * 0}. (7) 3X' � R (X)such thatX R*(X'). (14)
*
=

Definition 1[6]. Considering a subset X � U and a Remark. Let R be a binary relation on U, and let
binary relation R on U, Xis R-definable if and only if:
�R = {S � UIS E U fER and S * R.(0)}. (15)
3X' �X such thatX = UXEX!Rp(x) , (8)
We point out some insights of Defmition 2.
'<Ix E X',Rs(x) � X. (9) 1. If a set X � U is R-definable then X () R (0) 0.
*
=

Such a set X' is called a reference set of the R-defmable 2. Any R-defmable set X must be a member of the
setX family {R*(Y)IY � U}; consequently, Xis the union
Lemma 1 [6]. Considering a binary relation R on U, if a members of �R.
subset X� U is R-defmable set then the set 3. A subset X � U is R-definable if and only if
X" = {x E UI Rs(x) � X and Rp(x) � X} is the
·· R*(R (X) () (Wl) (X)) = X (16)
maximal reference set in the sense that X is a reference set * *
of Xand contains any reference set X'. v. AN ILLUSTRATIVE EXAMPLE
Notice [7] that
To illustrate what we have presented in the previous
X" = {x E UI Rs(x) �X and Rp(x) � X} sections, the following is an example using a reflexive (but
not symmetric or transitive) relation and a reflexive and
= R.(X) () (Wl).(X). (10) symmetric (but not transitive) relation.
Example 1. Consider a set U {Xl'X2,X3'x4} and a
=

binary
III. CHARACTERIZATION OF UPPER ApPROXIMATIONS
FOR BINARY RELATIONS R = {(Xl'Xl),(X2'X2),(X3'X3),(Xl'X2),(xvX3),
Motivated by the proof of an axiomatization of binary (X2'Xl),(X3'Xl)'(X3'X2),(X4'X3)}.
relations in [17], we obtain the following result.
Lemma 2. Let R be a binary relation on U. For each
R is neither reflexive nor symmetric. The R-successor
x E U, we have and R-predecessor sets are:

R*({x}) Rp(x). = (11) Rs(Xl) = {Xl'X2'X3}' Rp(xl) = {XvX2,X3}'

RS(X2) = {Xl,X2}, Rp(X2) = {XvX2,X3},
Proof: Let R be a binary relation on U, and letx E U. Rs(X3) = {Xl'X2'X3}' Rp(X3) = {XvX3,X4}'
According to (4), we have RS(X4) Rp(X4) 0.
= {X3}, =

R*({x}) = {y E UIRs(Y) () {x} * 0}. Accordingly, we have

= {y E Ulx E Rs(Y)} R.(0){x4},
=

= {y E UI(y,x) E R} U fER {{xvX3},{X2},{x4}},

=

= Rp(x). (12) �R {{Xl,X3}'{xz}},

=

Based on this lemma, (7) for a reflexive relation can be and we derive the lower and upper approximations as shown
easily extended to an arbitrary binary relation. in Table I.
Theorem 2. Considering a subset X � U and a binary In this example, the set X = {XVX3} E �R is not R­
relation R defined on U, we have defmable since it is not a member of{R*(y)IY � U}.
Let us consider the set X={Xl,X2,X3} in{R*(y)IY � U}.
UXEXRp(x) = {x E UIRs(x) () X * 0}. (13) Here Xis R-definable since

Proof: Let R be a binary relation on U, and letX E U. R*(R (X) () (Wl) (X)) =R*({xvX2,X3,X4} () {Xl'X2,X4})
* *
Using (4), (6) and Lemma 2, we have =R'({Xl,X2,X4}) {XVX2,X3}.
=

234
 TABLE T. THE LOWER AND UPPER APPROXIMATIONS

x R.(X) (R l).(X) R'(X) (R l)'(X)

C/J C/J fx ,} C/J C/J
{XL } C/J {X ,} {Xi,X2,X3 } {X ,X2,X3 }
fX2 1 C/J fx, l fx"x2,x, 1 fx ,X2 }
fx, } fx} fx, } fx"x"X, } fx,x7,x, }
{X4 } C/J {X,} C/J {X3 }
fx"x2 1 fX2 1 fx ,1 fx"xvx, l fx ,xvx, l
fx"x, } fx, } fx, } fx,X7,X"X, } fx,X7,X, }
{xvx, } C/J {x,} {x 'X2'X3'X4} {x 'x2'x3}
{x2,x3 1 {x l {x ,1 {x ,x2,x3,x4 1 {x ,x2,x3 1
fX7,X ,} C/J fx, } fX"X7,X, } fx ,X7,X, }
{X3'X ,} {x} {x,} {Xi,X3'X ,} {x,X2,X3 }
{x ,xvx, l {x ,x2,x"x, 1 {x"xvx , 1 {x ,x2,x"x4 1 {x ,xvx, l
fx ,x7,x. } fx7 } fx, } fx"x7,X, } fx,x7,x, }
{x 'x3'x4 } {x4 } {X3,X4 } {x ,X2,X3,X4 } {x ,X2,X3 }
{xvx"x4 1 fx l fx ,1 {x ,x2,x"x4 1 {x ,xvx, l
fX"X7,X"X, } fx,X7,X"X, } fx,X7,X"X, } fx,X7,X"X, } fx,X7,X, }

[5] T. Y. Lin, Granular computing on binary relations I: data mining and

VI. CONCLUSION neighborhood systems, In: A Skowron & L. Polkowski (eds.), Rough
sets and knowledge discovery, pp. 107-121, 1998, Physica-Verlag.
Generalized rough sets based on reflexive relations,
[6] R. Slowinski and D. Vanderpooten, "A generalized definition of
including tolerance (reflexive and symmetric) relations and rough approximations based on similarity", IEEE Transactions based
preorders (reflexive and transitive) relations, have immediate on Data and Knowledge Engineering, 12, 2000, 331-336.
applications to problems in incomplete information systems [7] Y. R. Syau and L. Jia, "Generalized rough sets based on reflexive
[19], i.e. to systems in which attribute values for objects may relations", Communications in Information and Systems, 12, 2012,
be unknown (missing, null). In such applications, 233-249.
Kryszkiewicz [20] introduced the use of a tolerance relation [8] C. Wang, C. Wu and D. Chen, "A systematic study on attribute
reduction with rough sets based on general binary relations",
to deal with the missing value case. Stefanowski and
Information Sciences, 178, 2008, 2237-2261.
Tsoukias [21] introduced the use of a reflexive and transitive
[9] H. P. Zhang, Y. Ouyang and Z. Wang, Note on "Generalized rough
relation to formalize the idea of absent value semantics. In sets based on reflexive and transitive relations", Information Sciences,
[22], Grzymala-Busse introduced the so-called characteristic 179, 2009, 471-473.
relation that is reflexive, but not necessarily symmetric, or [10] 1. Jarvinen and S. Jari Kortelainen, A note on definability in rough set
transitive. theory. In: R. De Caluwe, G. De Tr'e 1. Fodor, 1. Kacprzyk, S.
In [8], Wang et al. introduced and studied relation Zadro'zny, B. De Baets (Eds.), Current Issues in Data and Knowledge
information decision tables based on general binary relations. Engineering, Akademicka Oficyna Wydawnicza EXIT, 2004, pp.
272-277.
Definability of decision classes in relation information
[II] J. W. Grzymala-Busse and W. Rzasa, Definability and other
decision tables could be applied to study inconsistency properties of approximations for generalized indiscemibility relations.
concept of the relation information decision tables. Other In: J. F. Peters, & A. Skowron (Eds.), Transactions on Rough Sets XI,
projects will include exploring the unifying rough set 2010, pp. 14-39, Springer-Verlag.
theories via large scaled granular computing [14] to develop [12] D. Pei, "On definable concepts of rough set models", Information
new techniques in cloud computing. The idea is to use our Sciences, 177, 2007, 4230-4239.
space with binary relation together with a new composite [13] Y. R. Syau, L. Jia and E. B. Lin, "Definability of approximations in
rough set model [23], to process incomplete composite reflexive relations", Proceeding of the 2013 IEEE International
Conference on Granular Computing, 2013, pp. 276-280.
information and use some incremental algorithms to process
[14] T. Y. Lin, Y. Liu and W. Huang, Unifying Rough Set Theories via
big data on a distributed platform. It is conceivable to adopt
Large Scaled Granular Computing, Fundamenta Informaticae, 127,
the method in [23] to deal with some new composite rough 2013, 413-428.
set model. [IS] T. Y. Lin, Granular computing for binary relations II: rough set
representations and belief functions. In: A. Skowron, L. Polkowski
REFERENCES
(Eds.), Rough Sets and Knowledge Discovery, Physica-Verlag, 1998,
[1] Z. Pawlak, Rough sets: theoretical aspects of reasoning about data. 112-140.
Dordrecht: Kluwer Academic Publishers, 1991, Chapters I & 2. [16] 1. Jarvinen, S. Radeleczki and L. Veres, Rough sets determined by
[2] Z. Pawlak, "Rough sets", International Journal of Computer and quasiorders, Order 26, 2009, 337-355.
Information Science, II, 1982, 341-356. [17] T. Y. Lin, "Granular computing for binary relations: clustering and
[3] Z. Pawlak, "Rough set approach to knowledge-based decision support" axiomatic granular operators", Proceedings of the North American
European Journal of Operational Research, 99, 1997, 48-57. Fuzzy Information Processing Society, 2008, pp. 430-433.
[4] Z. Pawlak, J. W. Grzymala-Busse, R. Slowinski and W. Ziarko, [18] G. Liu and W. Zhu, "The algebraic structures of generalized rough set
"Rough set", Communications of the Association of Computing theory" Information Sciences, 178, 2008, 4105-4113.
Machinery, 38, 1995, 89-95.

235
[19] R. Slowinski and J. Stefanowski, Rough classification in incomplete [22] 1. W. Grzymala-Busse, Data with missing attribute values:
infonnation systems", Mathematical and Computer Modelling, generalization of indiscemibility relation and rule induction. In: J.F.
12(10/22), 1989, 1347-1357. Peters, A. Skowron, J. W. Grzymala-Busse, B. Kostek, R. W.
[20] Kryszkiewicz, "Rough set approach to incomplete information Swiniarski, M. S. Szezuka (Eds.), Transactions on Rough Sets I, 2004,
system", Information Sciences, 112, 1998, 39-49. pp. 78-95, Springer, Heidelberg.

[21] 1. Stefanowski and A. Tsoukias, "Incomplete infonnation tables and [23] S. Li, T. Li and J. Hu, "Update of approximations in composite
rough classification", Computational Intelligence, 17, 2001, 545-566. information systems", Know!. -Based Syst. 83, 2015, 138-148.

236