Complexity of Rule Sets Induced from Incomplete Data with

Attribute-concept Values and “Do Not Care” Conditions

Patrick G. Clark

and Jerzy W. Grzymala-Busse

1,2

Department of Electrical Engineering and Computer Science, University of Kansas, Lawrence, KS 66045, U.S.A.

Department of Expert Systems and Artiﬁcial Intelligence, University of Information Technology and Management,

35-225 Rzeszow, Poland

Keywords:

Data Mining, Rough Set Theory, Probabilistic Approximations, MLEM2 Rule Induction Algorithm,

Attribute-concept Values, “Do Not Care” Conditions.

Abstract:

In this paper we study the complexity of rule sets induced from incomplete data sets with two interpretations

of missing attribute values: attribute-concept values and “do not care” conditions. Experiments are conducted

on 176 data sets, using three kinds of probabilistic approximations (lower, middle and upper) and the MLEM2

rule induction system. The goal of our research is to determine the interpretation and approximation that

produces the least complex rule sets. In our experiment results, the size of the rule set is smaller for attribute-

concept values for 12 combinations of the type of data set and approximation, for one combination the size

of the rule sets is smaller for “do not care” conditions and for the remaining 11 combinations the difference

in performance is statistically insigniﬁcant (5% signiﬁcance level). The total number of conditions is smaller

for attribute-concept values for ten combinations, for two combinations the total number of conditions is

smaller for “do not care” conditions, while for the remaining 12 combinations the difference in performance

is statistically insigniﬁcant. Thus, we may claim that attribute-concept values are better than “do not care”

conditions in terms of rule complexity.

1 INTRODUCTION

Rough set theory has been applied to many areas

of data mining. Fundamental concepts of rough set

theory are standard lower and upper approximations.

In this paper we will use probabilistic approxima-

tions. A probabilistic approximation, associated with

a probability α, is a generalization of the standard ap-

proximation. For α = 1, the probabilistic approxima-

tion is reduced to the lower approximation; for very

small positive α, it is reduced to the upper approxi-

mation. Research on theoretical properties of prob-

abilistic approximations started from (Pawlak et al.,

1988) and then continued in many papers, see, e.g.,

(Pawlak and Skowron, 2007; Pawlak et al., 1988;

Sle¸zak and Ziarko, 2005; Yao, 2008; Yao and Wong,

1992; Ziarko, 2008).

Incomplete data sets may be analyzed using global

approximations such as singleton, subset and concept

(Grzymala-Busse, 2003; Grzymala-Busse, 2004a;

Grzymala-Busse, 2004b). Probabilistic approxima-

tions for incomplete data sets and based on an arbi-

trary binary relation were introduced in (Grzymala-

Busse, 2011). The ﬁrst experimental results using

probabilistic approximations were published in (Clark

and Grzymala-Busse, 2011).

For our experiments we use 176 incomplete data

sets, with two types of missing attribute values:

attribute-concept values (Grzymala-Busse, 2004c)

and “do not care” conditions (Grzymala-Busse, 1991;

Kryszkiewicz, 1995; Stefanowski and Tsoukias,

1999). Additionally, in our experiments we use three

types of approximations: lower, middle, and upper.

The middle approximation is the most typical proba-

bilistic approximation, with α = 0.5.

In (Clark and Grzymala-Busse, 2014), the results

indicate that rule set performance, in terms of error

rate, for both missing attribute value interpretations

is not signiﬁcantly different. As a result, given two

rule sets with the same error rate, the more desirable

would be the least complex, both for comprehension

and computation performance. Therefore, the main

objective of this paper is research on the complex-

ity of rule sets induced from data sets with attribute-

concept values and “do not care” conditions. Com-

plexity is deﬁned in terms of the number of rules

Clark P. and Grzymala-Busse J..

Complexity of Rule Sets Induced from Incomplete Data with Attribute-concept Values and "Do Not Care" Conditions.

DOI: 10.5220/0005003400560063

In Proceedings of 3rd International Conference on Data Management Technologies and Applications (DATA-2014), pages 56-63

ISBN: 978-989-758-035-2

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

and the number of rule conditions, with larger num-

bers indicating greater complexity. Our main result is

that the simpler rule sets are induced from data sets

in which missing attribute values are interpreted as

attribute-concept values.

Our secondary objective is to identify the approx-

imation (lower, middle or upper) that produces the

lowest rule complexity. Our conclusion is that all

three kinds of approximations do not differ signiﬁ-

cantly with respect to the complexity of induced rule

sets.

2 INCOMPLETE DATA

We assume that the input data sets are presented in

the form of a decision table. An example of a deci-

sion table is shown in Table 1. Rows of the decision

table represent cases, while columns are labeled by

variables. The set of all cases will be denoted by U.

In Table 1, U = {1, 2, 3, 4, 5, 6, 7, 8}. Independent

variables are called attributes and a dependent vari-

able is called a decision and is denoted by d. The set

of all attributes will be denoted by A. In Table 1, A =

{Education, Skills, Experience}. The value for a case

x and an attribute a will be denoted by a(x).

In this paper we distinguish between two interpre-

tations of missing attribute values: attribute-concept

values and “do not care” conditions. Attribute-

concept values, denoted by “−”, indicate that the

missing attribute value may be replaced by any speci-

ﬁed attribute value for a given concept. For example,

if a patient is sick with ﬂu, and if for other such pa-

tients the value of temperature is high or very-high,

then we will replace the missing attribute values of

temperature by values high and very-high, for details

see (Grzymala-Busse, 2004c). “Do not care” con-

ditions , denoted by “*”, mean that the original at-

tribute values are irrelevant, so we may replace them

by any attribute value, for details see (Grzymala-

Busse, 1991; Kryszkiewicz, 1995; Stefanowski and

Tsoukias, 1999). Table 1 presents an incomplete data

set affected by both attribute-concept values and “do

not care” conditions.

One of the most important ideas of rough set the-

ory (Pawlak, 1982) is an indiscernibility relation, de-

ﬁned for complete data sets. Let B be a nonempty

subset of A. The indiscernibility relation R(B) is a re-

lation on U deﬁned for x, y ∈ U as follows:

(x, y) ∈ R(B) if and only if ∀a ∈ B (a(x) = a(y)).

The indiscernibility relation R(B) is an equivalence

relation. Equivalence classes of R(B) are called ele-

mentary sets of B and are denoted by [x]

. A subset of

Table 1: A decision table.

Attributes Decision

Case Education Skills Experience Productivity

1 higher high − high

2 * high low high

3 secondary − high high

4 higher * high high

5 elementary high low low

6 secondary − high low

7 − low high low

8 elementary * − low

U is called B-deﬁnable if it is a union of elementary

sets of B.

The set X of all cases deﬁned by the same value

of the decision d is called a concept. For example,

a concept associated with the value low of the deci-

sion Productivity is the set {1, 2, 3, 4}. The largest

B-deﬁnable set contained in X is called the B-lower

approximation of X , denoted by appr

(X), and de-

ﬁned as follows

∪{[x]

| [x]

⊆ X },

while the smallest B-deﬁnable set containing X , de-

noted by appr

(X) is called the B-upper approxima-

tion of X, and is deﬁned as follows

∪{[x]

| [x ]

∩ X ̸=

0}.

For a variable a and its value v, (a, v) is called

a variable-value pair. A block of (a, v), denoted by

[(a, v)], is the set {x ∈ U | a(x) = v} (Grzymala-Busse,

1992).

For incomplete decision tables the deﬁnition of a

block of an attribute-value pair is modiﬁed in the fol-

lowing way.

• If for an attribute a there exists a case x such that

a(x) = −, then the corresponding case x should be

included in blocks [(a, v)] for all speciﬁed values

v ∈ V (x, a) of attribute a, where

V (x, a) =

= {a(y) | a(y) is speciﬁed, y ∈ U, d(y) = d(x )},

• If for an attribute a there exists a case x such that

a(x) = ∗, then the case x should be included in

blocks [(a, v)] for all speciﬁed values v of the at-

tribute a.

For the data set from Table 1, V (1, Experience) =

{low, high}, V (3, Skills) = {high},

ComplexityofRuleSetsInducedfromIncompleteDatawithAttribute-conceptValuesand"DoNotCare"Conditions

V (6, Skills) = {low, high}, V (7, Education) =

{elementary, secondary} and V (8, Experience) =

{low, high}.

For the data set from Table 1 the blocks of

attribute-value pairs are:

[(Education, elementary)] = {2, 5, 7, 8},

[(Education, secondary)] = {2, 3, 6, 7},

[(Education, higher)] = {1, 2, 4},

[(Skills, low)] = {4, 6, 7, 8},

[(Skills, high)] = {1, 2, 3, 4, 5, 6, 8},

[(Experience, low)] = {1, 2, 5, 8},

[(Experience, high)] = {1, 3, 4, 6, 7, 8}.

For a case x ∈ U and B ⊆ A, the characteristic set

(x) is deﬁned as the intersection of the sets K(x, a),

for all a ∈ B, where the set K(x, a) is deﬁned in the

following way:

• If a(x) is speciﬁed, then K(x, a) is the block

[(a, a(x))] of attribute a and its value a(x),

• If a(x) = −, then the corresponding set K(x, a)

is equal to the union of all blocks of attribute-

value pairs (a, v), where v ∈ V (x, a) if V (x, a) is

nonempty. If V (x, a) is empty, K(x, a) = U,

• If a(x) = ∗ then the set K(x, a) = U, where U is

the set of all cases.

For Table 1 and B = A,

(1) = {1, 2, 4},

(2) = {1, 2, 5, 8},

(3) = {3, 6},

(4) = {1, 4},

(5) = {2, 5, 8},

(6) = {3, 6, 7},

(7) = {6, 7, 8},

(8) = {2, 5, 7, 8}.

Note that for incomplete data there are a few

possible ways to deﬁne approximations (Grzymala-

Busse, 2003), we used concept approximations

(Grzymala-Busse, 2011) since our previous experi-

ments indicated that such approximations are most ef-

ﬁcient (Grzymala-Busse, 2011). A B-concept lower

approximation of the concept X is deﬁned as follows:

BX = ∪{K

(x) | x ∈ X, K

(x) ⊆ X },

while a B-concept upper approximation of the con-

cept X is deﬁned by:

BX = ∪{K

(x) | x ∈ X, K

(x) ∩ X ̸=

0} =

= ∪{K

(x) | x ∈ X}.

For Table 1, A-concept lower and A-concept upper

approximations of the concept {5, 6, 7, 8} are:

A{5, 6, 7, 8} = {6, 7, 8},

A{5, 6, 7, 8} = {2, 3, 5, 6, 7, 8}.

3 PROBABILISTIC

APPROXIMATIONS

For completely speciﬁed data sets a probabilistic ap-

proximation is deﬁned as follows

appr

(X) = ∪{[x] | x ∈ U, P(X | [x]) ≥ α},

α is a parameter, 0 < α ≤ 1, see (Grzymala-Busse,

2011; Grzymala-Busse and Ziarko, 2003; Pawlak

et al., 1988; Wong and Ziarko, 1986; Yao, 2008;

Ziarko, 1993). Additionally, for simplicity, the ele-

mentary sets [x]

are denoted by [x]. For discussion

on how this deﬁnition is related to the variable preci-

sion asymmetric rough sets see (Clark and Grzymala-

Busse, 2011; Grzymala-Busse, 2011).

Note that if α = 1, the probabilistic approximation

becomes the standard lower approximation and if α is

small, close to 0, in our experiments it is 0.001, the

same deﬁnition describes the standard upper approxi-

mation.

For incomplete data sets, a B-concept probabilis-

tic approximation is deﬁned by the following formula

(Grzymala-Busse, 2011)

∪{K

(x) | x ∈ X, Pr(X|K

(x)) ≥ α}.

For simplicity, we will denote K

(x) by K(x)

and the A-concept probabilistic approximation will be

called a probabilistic approximation.

For Table 1 and the concept X = [(Productivity,

low)] = {5, 6, 7, 8}, there exist three distinct three

distinct probabilistic approximations:

appr

1.0

(

{

}

) =

{

}

appr

0.75

({5, 6, 7, 8}) = {2, 5, 6, 7, 8},

and

appr

0.001

({5, 6, 7, 8}) = {2, 3, 5, 6, 7, 8}.

The special probabilistic approximations with the

parameter α = 0.5 will be called a middle approxima-

tion.

4 EXPERIMENTS

Our experiments are based on eight data sets available

from the University of California at Irvine Machine

Learning Repository, see Table 2.

For every data set a set of templates is created

by incrementally replacing a percentage of existing

speciﬁed attribute values (at a 5% increment) with

attribute-concept values. Thus, we started each se-

ries of experiments with no attribute-concept values,

DATA2014-3rdInternationalConferenceonDataManagementTechnologiesandApplications

0 5 10 15 20 25 30

Rule count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 1: Size of the rule set for the Bankruptcy data set.

100

120

0 10 20 30 40

Rule count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 2: Size of the rule set for the Breast cancer data set.

0 5 10 15 20 25 30 35

Rule count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 3: Size of the rule set for the Echocardiogram data

set.

then we changed 5% of speciﬁed values to attribute-

concept values, then we changed an additional 5% of

speciﬁed values to attribute-concept values, etc., un-

til at least one entire row of the data set is full of

attribute-concept values. Then three attempts were

made to change the conﬁguration of new attribute-

concept values and either a new data set with an extra

5% of attribute-concept is created or the process is

0 10 20 30 40 50

Rule count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 4: Size of the rule set for the Hepatitis data set.

0 10 20 30 40 50 60 70

Rule count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 5: Size of the rule set for the Image segmentation

data set.

0 5 10 15 20 25 30 35

Rule count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 6: Size of the rule set for the Iris data set.

terminated. Additionally, the same formed templates

are edited for further experiments by replacing each “-

”, representing attribute-concept values with “*”, rep-

resenting “do not care” conditions.

For any data set there is some maximum for the

percentage of missing attribute values. For example,

for the bankruptcy data set, it is 35%. Hence, for the

bankruptcy data set, there exist seven data sets with

ComplexityofRuleSetsInducedfromIncompleteDatawithAttribute-conceptValuesand"DoNotCare"Conditions

0 10 20 30 40 50 60 70

Rule count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 7: Size of the rule set for the Lymphography data set.

0 10 20 30 40 50 60

Rule count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 8: Size of the rule set for the Wine recognition data

set.

0 5 10 15 20 25 30

Condition count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 9: Number of conditions for the Bankruptcy data set.

attribute-concept values and seven data sets with “do

not care” conditions, for a total of 15 data sets (the ad-

ditional data set is complete, with no missing attribute

values). In a similar process for the breast cancer,

echocardiogram, hepatitis, image segmentation, iris,

lymphography and wine recognition data sets we cre-

ated 19, 17, 25, 29, 15, 29, and 27 data sets. Hence

the total number of data sets is 176.

100

150

200

250

300

350

400

0 10 20 30 40

Condition count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 10: Number of conditions for the Breast cancer data

set.

0 5 10 15 20 25 30 35

Condition count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 11: Number of conditions for the Echocardiogram

data set.

100

120

0 10 20 30 40 50

Condition count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 12: Number of conditions for the Hepatitis data set.

Results of our experiments are presented in Fig-

ures 1–16.

DATA2014-3rdInternationalConferenceonDataManagementTechnologiesandApplications

100

150

200

250

300

350

400

450

0 10 20 30 40 50 60

Condition count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 13: Number of conditions for the Image segmenta-

tion data set.

100

0 5 10 15 20 25 30

Condition count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 14: Number of conditions for the Iris data set.

100

120

140

160

180

0 10 20 30 40 50 60

Condition count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 15: Number of conditions for the Lymphography

data set.

5 DISCUSSION

First we compare two interpretations of missing at-

tribute values, attribute-concept values and “do not

care” conditions with respect to the rule set size. For

100

120

140

160

0 10 20 30 40 50 60

Condition count

Missing attributes (%)

Lower, -

Middle, -

Upper, -

Lower, *

Middle, *

Upper, *

Figure 16: Number of conditions for the Wine recognition

data set.

Table 2: Data sets used for experiments.

Data set Number of

cases attributes concepts

Bankruptcy 66 5 2

Breast cancer 277 9 2

Echocardiogram 74 7 2

Hepatitis 155 19 2

Image segmentation 210 19 7

Iris 150 4 3

Lymphography 148 18 4

Wine recognition 178 13 3

every data set type, separately for lower, middle and

upper approximations, the Wilcoxon matched-pairs

signed rank test is used with a 5% level of signiﬁcance

two-tailed test. With eight data set types and three ap-

proximation types, the total number of combinations

is 24.

For 12 combinations the rule set size is smaller for

attribute-concept values: bankruptcy data set with all

three types of approximations, hepatitis data set with

middle and upper approximations, image segmenta-

tion data set with middle and upper approximations,

iris data set with all three types of approximations

and wine recognition data set with middle and upper

approximations. For one combination, breast cancer

data set with middle approximations, the size of the

rule set is smaller for “do not care” conditions. For

the remaining 11 combinations, the difference in the

rule set size between attribute-concept values and “do

not care” conditions is insigniﬁcant. Therefore there

is strong evidence that attribute-concept values pro-

vide for smaller rule set sizes than “do not care” con-

ditions.

Similarly, for the total number of conditions in a

rule set, in ten combinations this number is smaller for

ComplexityofRuleSetsInducedfromIncompleteDatawithAttribute-conceptValuesand"DoNotCare"Conditions

attribute-concept values: bankruptcy data set with all

three types of approximations, echocardiogram data

set with all three types of approximations, lymphog-

raphy data set with upper approximations and wine

recognition data set with all three types of approx-

imations. For two combinations the total number

of conditions in the rule set is smaller for “do not

care” conditions: breast cancer data set with middle

approximations and image recognition data set with

lower approximations. For the remaining 12 combi-

nations the difference in the total number of condi-

tions in rule sets between attribute-concept values and

“do not care” conditions is insigniﬁcant. Thus there

is evidence that attribute-concept values provide for

a smaller total number of conditions in rule sets than

“do not care” conditions.

Next, for a given interpretation of missing at-

tribute values we compare all three types of approx-

imations in terms of the rule set size and the total

number of conditions in the rule set. For all eight

types of data sets, we compare lower approximations

with middle and upper approximations, and middle

approximations with upper approximations. This ex-

periment setup results in a total of 24 combinations

and the Friedman Rank Sums test, with 5% signiﬁ-

cance level, is used.

The rule set size is smaller for lower approxi-

mations than for upper approximations in four com-

binations of the type of data set and type of miss-

ing attribute value: hepatitis data set with attribute-

concept values, image segmentation data set with both

attribute-concept values and “do not care” conditions

and wine recognition data set with “do not care” con-

ditions. The rule set size is smaller for lower approx-

imations than for middle approximations for three

combinations of the data set type and missing attribute

value type: image segmentation data set with both

attribute-concept values and “do not care” conditions

and hepatitis data set with attribute-concept values.

The rule set size is smaller for upper approximations

than for lower approximations in three combinations:

bankruptcy data set with “do not care” conditions, iris

data set with “do not care” conditions and lymphog-

raphy data set with attribute-concept values. Finally,

the rule set size is smaller for upper approximations

than for middle approximations in one combination:

breast cancer data set with attribute-concept values.

In the remaining 13 combinations the difference be-

tween all three approximations is insigniﬁcant. There

is weak evidence that lower approximations might be

better than the remaining approximations.

The total number of conditions in a rule set is

smaller for lower approximations than for upper ap-

proximations in three combinations of data set type

and missing attribute value type: hepatitis data set

with attribute-concept values and image segmentation

data set with both attribute-concept values and “do

not care” conditions. The total number of conditions

in a rule set is smaller for lower approximations than

for middle approximations in one combination: image

segmentation data set with “do not care” conditions.

The total number of conditions in a rule set is smaller

for middle approximations than for lower approxima-

tions in one combination: lymphography data set with

attribute-concept values. The total number of condi-

tions in a rule set is smaller for upper approximations

than for lower approximations in three combinations:

bankruptcy data set with “do not care” conditions, iris

data set with “do not care” conditions and lymphogra-

phy

data set with attribute-concept values. Finally, the

total number of conditions in a rule set is smaller for

upper approximations than for middle approximations

in one combination: lymphography data set with the

attribute-concept values. In the remaining 15 combi-

nations the difference between all three approxima-

tions is insigniﬁcant. Practically speaking, we do not

have enough evidence to tell which approximation is

the best.

6 RELATED WORK

Rough set concepts and the indiscernibility relation

are introduced in (Pawlak, 1982) with additional re-

search explaining theoretical concepts of probabilistic

approximations in (Pawlak et al., 1988). Further re-

search was conducted in the area of probabilistic types

of approximations in other efforts, comparing them

to deterministic approaches and studying other exten-

sions in (Pawlak and Skowron, 2007; Pawlak et al.,

1988;

Sle¸zak and Ziarko, 2005; Yao, 2008; Yao and

Wong, 1992; Ziarko, 2008).

In (Grzymala-Busse, 2003; Grzymala-Busse,

2004a; Grzymala-Busse, 2004b), three global approx-

imations are deﬁned: singleton, subset and concept.

The papers also include a study of these approxima-

tions with incomplete data. In our work, concept ap-

proximations are used in conjunction with probabilis-

tic approximations on incomplete data sets. These

concepts were introduced in (Grzymala-Busse, 2011)

and include deﬁnitions of B-concept probabilistic ap-

proximations with discussions on how the deﬁnition

is related to variable precision asymmetric rough sets.

In addition, the ﬁrst experimental results are studied

in (Clark and Grzymala-Busse, 2011).

DATA2014-3rdInternationalConferenceonDataManagementTechnologiesandApplications

7 CONCLUSIONS

As follows from our experiments, there is evidence

that the rule set size is smaller for the attribute-

concept interpretation of missing attribute values than

for the “do not care” condition interpretation. The to-

tal number of conditions in rule sets is also smaller for

attribute-concept interpretation of missing attribute

values. Thus we may claim attribute-concept values

are better than “do not care” conditions as an inter-

pretation of a missing attribute value in terms of rule

complexity.

Furthermore, all three kinds of approximations

(lower, middle and upper) do not differ signiﬁcantly

with respect to the complexity of induced rule sets.

Future direction this work might take is an inves-

tigation of other interpretations of missing attribute

values and a comparison of the complexity of the rule

sets produced.

REFERENCES

Clark, P. G. and Grzymala-Busse, J. W. (2011). Experi-

ments on probabilistic approximations. In Proceed-

ings of the 2011 IEEE International Conference on

Granular Computing, pages 144–149.

Clark, P. G. and Grzymala-Busse, J. W. (2014). Mining in-

complete data with attribute-concept values and “do

not care” conditions. In Proceedings of the 9th Inter-

national Conference on Hybrid Artiﬁcial Intelligence

Systems, pages 146–167.

Grzymala-Busse, J. W. (1991). On the unknown attribute

values in learning from examples. In Proceedings

of the ISMIS-91, 6th International Symposium on

Methodologies for Intelligent Systems, pages 368–

377.

Grzymala-Busse, J. W. (1992). LERS—a system for learn-

ing from examples based on rough sets. In Slowinski,

R., editor, Intelligent Decision Support. Handbook of

Applications and Advances of the Rough Set Theory,

pages 3–18. Kluwer Academic Publishers, Dordrecht,

Boston, London.

Grzymala-Busse, J. W. (2003). Rough set strategies to data

with missing attribute values. In Workshop Notes,

Foundations and New Directions of Data Mining, in

conjunction with the 3-rd International Conference on

Data Mining, pages 56–63.

Grzymala-Busse, J. W. (2004a). Characteristic relations for

incomplete data: A generalization of the indiscerni-

bility relation. In Proceedings of the Fourth Interna-

tional Conference on Rough Sets and Current Trends

in Computing, pages 244–253.

Grzymala-Busse, J. W. (2004b). Data with missing attribute

values: Generalization of indiscernibility relation and

rule induction. Transactions on Rough Sets, 1:78–95.

Grzymala-Busse, J. W. (2004c). Three approaches to miss-

ing attribute values—a rough set perspective. In Pro-

ceedings of the Workshop on Foundation of Data Min-

ing, in conjunction with the Fourth IEEE International

Conference on Data Mining, pages 55–62.

Grzymala-Busse, J. W. (2011). Generalized parameterized

approximations. In Proceedings of the RSKT 2011,

the 6-th International Conference on Rough Sets and

Knowledge Technology, pages 136–145.

Grzymala-Busse, J. W. and Ziarko, W. (2003). Data mining

based on rough sets. In Wang, J., editor, Data Mining:

Opportunities and Challenges, pages 142–173. Idea

Group Publ., Hershey, PA.

Kryszkiewicz, M. (1995). Rough set approach to incom-

plete information systems. In Proceedings of the

Second Annual Joint Conference on Information Sci-

ences, pages 194–197.

Pawlak, Z. (1982). Rough sets. International Journal of

Computer and Information Sciences, 11:341–356.

Pawlak, Z. and Skowron, A. (2007). Rough sets: Some

extensions. Information Sciences, 177:28–40.

Pawlak, Z., Wong, S. K. M., and Ziarko, W. (1988). Rough

sets: probabilistic versus deterministic approach. In-

ternational Journal of Man-Machine Studies, 29:81–

95.

Sle¸zak, D. and Ziarko, W. (2005). The investigation of the

bayesian rough set model. International Journal of

Approximate Reasoning, 40:81–91.

Stefanowski, J. and Tsoukias, A. (1999). On the exten-

sion of rough sets under incomplete information. In

Proceedings of the RSFDGrC’1999, 7th International

Workshop on New Directions in Rough Sets, Data

Mining, and Granular-Soft Computing, pages 73–81.

Wong, S. K. M. and Ziarko, W. (1986). INFER—an adap-

tive decision support system based on the probabilistic

approximate classiﬁcation. In Proceedings of the 6-th

International Workshop on Expert Systems and their

Applications, pages 713–726.

Yao, Y. Y. (2008). Probabilistic rough set approxima-

tions. International Journal of Approximate Reason-

ing, 49:255–271.

Yao, Y. Y. and Wong, S. K. M. (1992). A decision theoretic

framework for approximate concepts. International

Journal of Man-Machine Studies, 37:793–809.

Ziarko, W. (1993). Variable precision rough set model.

Journal of Computer and System Sciences, 46(1):39–

59.

Ziarko, W. (2008). Probabilistic approach to rough

sets. International Journal of Approximate Reason-

ing, 49:272–284.

ComplexityofRuleSetsInducedfromIncompleteDatawithAttribute-conceptValuesand"DoNotCare"Conditions