Beruflich Dokumente
Kultur Dokumente
INTRODUCTION
1.1
MEASURE:
to certain
subsets of a set X(Ref.1 [4]). It must assign 0 to the empty set and be countably additive: the
measure of a large subset that can be decomposed into a countable number of smaller disjoint
subset, is the sum of the measures of the smaller subsets by(Ref.1 [8]). In general, if one wants
to associate a consistent size to each subset of a given set while satisfying the other axioms of a
1
measure, one only finds trivial examples like the counting measure(Ref.1 [2]). This problem was
resolved by defining measure only on a sub-collection of all subsets; the so called measurable
subsets, which are required to form a
Definition:
: [ ,+]
Non-negativity:
( E) 0
(b)
( )=0
(c)
for all
- additivity):
( i I Ei )= ( E i ) (1.1)
i I
One may require that at least one set E has finite measure. Then the null set automatically has
measure zero because of countable additivity, because
i=0
i=0
( E )= ( E .. )= ( E ) + ( ) ( ) ,(1.2)
is finite if and only if the empty set has measure zero. If only the second and third conditions of
the definition of measure above are met, and
Properties
2
Several further properties can be derived from the definition of a countably additive measure
by(Ref.1 [6, 8, 9]).
(a)
Monotonicity: A measure
is monotonic: If
E1
and
E2
with
E1 E 2
(b)
then,
( E1 ) ( E2 )
(1.3)
A measure
Ei ) ( Ei ).(1.4)
i=1
A measure is continuous from below: If E 1, E2, E3, . are measurable sets and En is a subset
of En+1 for all n, then the union of the sets En is measurable, and
i=1
( Ei ) =lim ( Ei) .(1.5)
i
(c)
A measure
is continuous from above: If E1, E2, E3, . are measurable sets and En+1 is a
subset of En for all n, then the intersection of the sets En is measurable; furthermore, if at least one
of the En has finite measure, then
i=1
( Ei ) =lim ( Ei) .(1.6)
i
This property is false without the assumption that at least one of the E n has finite measure. For
instance, for each
n N , let
En= R
intersection is empty.
1.1.3. Some measures are defined here by (Ref.1 [2, 5, 8, 9])
I
Sigma-finite measure: A measure space (X, , ) is called finite if (X) is a finite real
number. It is called
finite. Consider the closed intervals [k, k+1] for all integers k; there are countably many such
intervals, each has measure 1, and their union is the entire real line. Alternatively, consider the
real numbers with the counting measure, which assigns to each finite set of real numbers i.e.
number of points in the set. This measure space is not
measure contains only finitely many points, and it would take uncountably many such set to
cover the entire real line. The
properties;
topological spaces. They can be also thought of as a vague generalization of the idea that a
measure space may have uncountable measure.
II
. A subset of a
null set is called a negligible set. A negligible set need not be measurable, but every measurable
negligible set is automatically a null set. A measure is called complete measure if every
negligible set is measurable.
algebra of subsets Y
which differ by a negligible set from a measurable set X, that is, such that the symmetric
difference of X and Y is contained in a null set.
III
any family
<
and
[ a ] X
[ b] ( X )= ( X ) .
Note that the second condition is equivalent to the statement that the ideal of null sets is
complete. Some other important measures are listed here by (Ref.1[1, 3, 8])
IV
with values in the (signed) real numbers is called a signed measure, while such a function with
values in the complex numbers is called a complex measure. Measures that take values in Banach
spaces have been studied extensively (by Ref.1 [4]). A measure that takes values in the set of selfadjoint projections on a Hilbert space is called a projection-valued measure; these are used in
functional analysis for the spectral theorem. When it is necessary to distinguish the usual
measures, which take non-negative values from generalizations, the term positive measure is
used. Positive measures are closed under conical combination but not general linear combination,
while signed measures are the linear closure of positive measures(Ref.1 [2]).
Another generalization is the finitely additive measure, which are sometimes called contents.
This is the same as a measure except that instead of requiring countable additivity we require
only finite additivity. Finite additive measures are connected with notions such as Banach limits,
the dual of
L and the Stone-Cech compactification (by Ref.1 [8]). All these are linked in one
1.2.
FUZZY:
Fuzzy mathematics forms a branch of mathematics related to fuzzy set theory and fuzzy logic. It
started in 1965 after the publication of Lotfi Asker Zadehs seminal work Fuzzy Sets (by
Ref.2[1]). A fuzzy subset A of a set X is a function
L={0,1} . More
generally, one can use a complete lattice L in a definition of a fuzzy subset A (Ref.2 [2]).
The evolution of the fuzzification of mathematical concepts can be broken down into three
stages: (by Ref.2 [10])
(a)
(b)
the explosion of the possible choices in the generalization process during the eighties,
(c)
and union
A B
x , y X , A ( xy ) min ( A ( x ) , A ( y ) ) .
[11]).
A fuzzy is a concept of which the meaningful content, value, or boundaries of application can
vary considerably according to context or conditions, instead of being fixed once and for all
(Ref.2[3]). This generally means the concept is vague, lacking a fixed, precise meaning, without
however being meaningless altogether (Ref.2[4]). It has a meaning, or multiple meaning. But
these can become clearer only through further elaboration and specification, including a closer
definition of the context in which they are used. Fuzzy concepts lack clarity and are difficult to
test or operationalize (Ref.2[5]).
In logic, fuzzy concepts are often regarded as concepts which in their application, or formally
speaking, are neither completely true nor completely false, or which are partly true or partly false;
7
they are ideas which require further elaboration, specification or qualification to understand their
applicability (Ref.2 [7]) (the conditions under which they truly make sense).
In mathematics and statistics, a fuzzy variable (such as the temperature, hot or cold) is a
value which could lie in a probable range defined by quantitative limits or parameters, and which
can be usefully described with imprecise categories (such as high, medium of low).
In mathematics and computer science, the gradations of applicable meaning of a fuzzy concept
are described in terms of quantitative relationships defined by logical operators. Such an
approach is sometimes called degree-theoretic semantics by logicians and philosophers (Ref.2
[6]), but more usual term is fuzzy logic or many-valued logic. The basic idea is, that a real
number is assigned to each statement written in a language, within a range from 0 to 1, where 1
means that the statement is completely true, and 0 means that the statement is completely false,
while values less than 1 but greater than 0 represent that the statements are partly true, to a
given, quantifiable extent. This makes it possible to analyze a distribution of statement for their
truth-content, identify data patterns, make inferences and predictions, and model how processes
operate.
Fuzzy reasoning (i.e. reasoning with graded concepts) has many practical uses (by Ref.2 [12]). It
is nowadays widely used in the programming of vehicle and transport electronics, household
appliances, video games, language filters, robotics, and various kinds of electronic equipment
used for pattern recognition, surveying and monitoring(such as radars). Fuzzy reasoning is also
used in artificial intelligence and virtual intelligence research.(Ref.2 [13]) Fuzzy risk scores are
used by project managers and portfolio managers to express risk assessments. Ref.2 [14].
A fuzzy set can be defined mathematically by assigning to each possible individual in the
universe of discourse a value representing its grade of membership in the fuzzy set. This grade
corresponds to the degree to which that individual is similar or compatible with the concept
represented by the fuzzy set. Thus, individuals may belong in the fuzzy set to a greater or lesser
degree as indicated by a larger or smaller membership grade. As already mentioned, these
membership grades are very often represented by real-number values ranging in the closed
interval between 0 and 1(by Ref.2 [9] on page 4-5). Thus, a fuzzy set representing our concept of
sunny might assign a degree of membership of 1 to a cloud cover of 0%, 0.8 to a cloud cover of
8
20%, 0.4 to a cloud cover of 30%, and 0 to a cloud cover of 75%. These grades signify the degree
to which each percentage of cloud cover approximate our subjective concept of sunny, and the set
itself models the semantic flexibility inherent in such a common linguistic term. Because full
membership and full non-membership in the fuzzy set can still be indicated by the values of 1 and
0, respectively, we can consider the concept of a crisp set to be a restricted case of the more
general concept of a fuzzy set for which only these two grades of membership are allowed( Ref.2
[9]).
1.3.
UNCERTAINTY
Fuzzy concepts can generate uncertainty because they are imprecise (especially if they refer to a
process in motion or a process of transformation where something is in the process of turning
into something else). In that case, they do not provide a clear orientation for action or decisionmaking, reducing fuzziness, perhaps by applying fuzzy logic, would generate more certainty
(Ref.2 [7]). A fuzzy concept may indeed provide more security, because it provides a meaning for
something when an exact concept is unavailable which is better than not being able to denote it
at all (Ref.2 [8]).
It is generally agreed that an important point in the evolution of the modern concept of
uncertainty was the publication of a seminal paper by (Ref.2 [1]), even though some ideas
presented in the paper were envisioned some few years earlier by (Ref.2 [15]). In this paper,
Zadeh introduced a theory whose objects fuzzy sets are sets with boundaries that are not
precise. The membership in a fuzzy set is not a matter of affirmation or denial, but rather a matter
of a degree (Ref.2 [9])
The significance of (Ref.2 [1]) was that it challenged not only probability theory as the sole agent
for uncertainty, but the very foundations upon which probability theory is based: Aristotelian
two-valued logic. When A is a fuzzy set and x is a relevant object, the proposition x is a member
of A is not necessarily either true or false, as required by two-valued logic, but it may be true
only to some degree, the degree to which x is actually a member of A. it is most common, but not
required, to express degrees of membership in fuzzy sets as well as degree of truth of the
associated propositions by numbers in the closed unit interval [0,1]. The extreme values in this
9
interval, 0 and 1, then represent, respectively, the total denial and affirmation of the membership
in a given fuzzy set as well as the falsity and truth of the associated proposition.
For instance, suppose we are trying to diagnose an ill patient. In simplified terms, we may be
trying to determine whether this patient belongs to the set of people with, say, pneumonia,
bronchitis, emphysema, or a common cold. A physical examination may provide us with helpful
yet inconclusive evidence (by Ref.2 [9] on page 179). For example, we might assign a high
values, say 0.75, to our best guess, bronchitis, and a lower value to the other possibilities, such as
0.45 for the set consisting of pneumonia and emphysema and 0 for a common cold. These values
reflect the degrees to which the patients symptoms provide evidence for the individual diseases
or set of diseases; their collection constitutes a fuzzy measure representing the uncertainty
associated with several well-defined alternatives. It is important to realize that this type of
uncertainty, which results from information deficiency, is fundamentally different from fuzziness,
which results from the lack of sharp boundaries.
Uncertainty-based information was first conceived in terms of classical set theory and in terms of
probability theory. The term information theory has almost invariably been used to a theory based
upon the well known measure of probabilistic uncertainty established by (Ref.2 [16]). Research
on a broader conception of uncertainty-based information, liberated from the confines of classical
set theory and probability theory, began in the early eighties. The name generalized information
theory was coined for a theory based upon this broader conception.
The ultimate goal of generalized information theory is to capture properties of uncertainty-based
information formalized within any feasible mathematical framework. Although this goal has not
been fully achieved as yet, substantial progress has been made in this direction. In addition to
classical set theory and probability theory, uncertainty-based information is now well understood
in fuzzy set theory, possibility theory and evidence theory.
Three types of uncertainty are now recognized in the five theories, in which measurement of
uncertainty is currently well established. These three uncertainty types are: nonspecificity (or
imprecision), which is connected with sizes (cardinalities) of relevant sets of alternatives;
fuzziness (or vagueness), which results from imprecise boundaries of fuzzy sets; and strife (or
discord), which expresses conflicts among the various sets of alternatives.
10
1.3.1
Measurement
of
uncertainty
and
associated
information was first conceived in terms of classical set theory. It was shown by (Ref.2 [17]) that
using a function from the class of functions
U ( A )=c . log b|A|,
where
(1.7)
| A| denotes the cardinality of a finite nonempty set A, and b, c are positive constants
(b>1, c >0) , is the only sensible way to measure the amount of uncertainty associated with a
finite set of possible alternatives. Each choice of values of the constants b and c determines the
unit in which uncertainty is measured. When b=2 and c=1, which is most common choice,
uncertainty is measured in bits, and obtain
U ( A )=log 2| A|.
(1.8)
One bit of uncertainty is equivalent to the total uncertainty regarding the truth or falsity of one
proposition. The set function U defined by equation (1.8) be called a Hartley function. Its
uniqueness as a measure of uncertainty associated with sets of alternatives can also be proven
axiomatically.
A natural generalization of the Hartley function from classical set theory to fuzzy set theory was
proposed in the early eighties under the name U- uncertainty. For any nonempty fuzzy set A
defined on a finite universal set X, the generalized Hartley function (Ref.2[17]) has the form
1
U ( A )=
h( A)
where
h (A )
log 2||d
(1.9)
11
preceding
of a given
-cut. Fuzzy sets are equal when normalized have the same nonspecificity measured
not crisp sets) is fuzziness (or vagueness). In general, a measure of fuzziness is a function
f : P(X) R ,
where P(X) denotes the set of all fuzzy subsets of X (fuzzy power set). For each
fuzzy set A, this function assign a nonnegative real number f(A) that express the degree to which
the boundary of A is not sharp.(Ref.2 [9])
One way is to measure fuzziness of any set A by a metric distance between its membership grade
function and the membership grade function (or characteristic function) of the nearest crisp set.
(Ref.2 [1]) Even when committing to this conception of measuring fuzziness, the measurement is
not unique. To make it unique, we have to choose a suitable distance function.
Another way of measuring fuzziness, which seems more practical as well as more general, is to
view the fuzziness of a set in terms of the lack of distinction between the set and it complement.
Indeed, it is precisely the lack of distinction between the sets and their complements that
distinguishes fuzzy sets from crisp sets.(Ref.2 [1]) The less a set differs from its complement, the
fuzzier it is. Let us restrict our discussion to this view of fuzziness, which is currently
predominant in the literature.
1.3.3. Strife (or discord):
evidential claims has been far more controversial. Although it is generally agreed that a measure
of this type of uncertainty must be a generalization of the well-established Shannon entropy
(Ref.2[17]) from probability theory, which all collapse to the Shannon entropy within the domain
of probability theory, is the right generalization. As argued in the companion paper (Ref.2 [18]),
either of these measures is deficient in some trails. To overcome these deficiencies, another
measure was prepared in the companion paper(Ref.2 [18]), which is called a measure of
discord. This measure is expressed by a function D and defined by the formula,
12
D ( m) = m ( A ) log2 1 m ( B )
A F
B F
|B A|
(1.10)
|B|
The rationale for choosing this function is explained as follows. The term,
Con ( A ) = m(B)
B F
|B A|
( 1.11)
|B|
In equation (1.10) expresses the sum of conflicts of individual evidential claim m(B) for all
B F
with respect to the evidential claim m(A) focusing on a particular set A; each individual
B A
is violated.
increasing with Con (A) and, consequently, it represents the same quantity as Con(A), but on the
logarithmic scale. The use of the logarithmic scale is motivated in the same way as in the case of
the Shannon entropy (Ref.2 [17]).
Function D is clearly a measure of the average conflict among evidential claims within a given
body of evidence. Consider, as an example, incomplete information regarding the age of a person
X. Assume that the information is expressed by two evidential claims pertaining to the age of X:
X is between 15 and 17 years old with degree m(A), where A = {15, 17}, and X is a teenager
with degree m(B), where B = {13, 19}. Clearly, the weaker second claim does not conflict with
the stronger first claim. Assume that
focusing on B is not implied by the claim focusing on A and, consequently, m(B) dose conflict
with m(A) to a degree proportional to number of elements in A that are not covered by B. This
conflict is not captured by function Con since |B A|=0
in this case.
It follows from these observations that the total conflict of evidential claims within a body of
evidence {F, m} with respect to a particular claim m(A) should be expressed by function
13
CON ( A )= m ( B )
B F
| AB|
(1.12)
|A|
rather than function Con given by equation (1.11). Replacing Con (A) with CON (A), we obtain a
new function, which is better justified as a measure of conflict in evidence theory than function
D. This new function, which is called strife and denoted by S, is defined by the form,
S ( m) = m( A)log 2 1 m( B)
A F
B F
| AB|
.(1.13)
| A|
1=r 1 r 2 r n
[ ]
i
j=1
r j ,(1.14)
Where U(r) is the measure of possibilistic nonspecificity (U-uncertainty). The maximum value of
possibilistic strife, given by equation (1.14), depends on n in exactly the same way as the
maximum value of possibilistic discord: it increases with n and converges to a constant, estimated
as 0.892 as
(Ref.2 [19]). However the possibility distributions for which the maximum
of possibilistic strife are obtained (one for each value of n) are different from those for
possibilistic discord. This property and the intuitive justification of function S make this function
better candidate for the entropy-like measure in Dempster-shafer theory (DST) than any of the
previously considered functions.
1.3.4. Applications of Measures of Uncertainty: Uncertainty
measures
becomes
well
justified, they can be used in many other contexts as for managing uncertainty and the associated
information. For example, they can be used for extrapolating evidence, assessing the strength of
relationship between given groups of variables, assessing the influence of given input variables
on given output variables, measuring the lost of information when a system is simplified, and the
like. In many problem situations, the relevant measures of uncertainty are applicable only in their
14
conditional or relative terms. The use of relevant uncertainty measures is as broad as the use of
any relevant measuring instrument.(Ref.2[24]) Three basic principles of uncertainty were
developed to guide the use of uncertainty measures in different situations. These principles are: a
principle of minimum uncertainty, a principle of maximum uncertainty, and a principle of
uncertainty invariance.
The principle of minimum uncertainty is an arbitration principle. It is used for narrowing down
solutions in various systems problems that involve uncertainty. It guides the selection of
meaningful alternatives from possible solutions of problems in which some of the initial
information is inevitably lost but in different solutions it is lost in varying degrees. The principle
states that we should accept only those solutions with the least loss of information i.e. whose
uncertainty is minimal. Application of the principle of minimum uncertainty is the area of
conflict-resolution problems. For some development of this principle see(Ref.2 [23]). For
example when we integrate several overlapping models into one larger model, the models may be
locally inconsistent. It is reasonable to require that each of the models be appropriately adjusted
in such a way that the overall model becomes consistent. It is obvious that some information
contained in the given models is inevitably lost by these adjustments. This is not desirable.
Hence, we should minimize this lost of information. That is, we should accept only those
adjustments for which the total increase of uncertainty is minimal.
The principle of maximum uncertainty is essential for any problem that involves applicative
reasoning. Applicative reasoning is indispensable to science in a variety of ways. For example,
when we want to estimate microstates from the knowledge of relevant microstates and partial
information regarding the microstates, we must resort to applicative reasoning. Applicative
reasoning is also common and important in our daily life where the principle of maximum
uncertainty is not always adhered to. Its violation leads almost invariable to conflicts in human
communication, as well expressed by (Ref.2 [26]) ..whenever you find yourself getting angry
about a difference in opinion, be on your guard; you will probably find, on examination, that your
belief is getting beyond what the evidence warrants. This is reasoning in which conclusions are
not entailed in the given premises. The principle may be expressed by the following requirement:
in any applicative inference, use all information available, but make sure that no additional
information is unwittingly added. The principle of maximum uncertainty is applicable in
15
situations in which we need to go beyond conclusions entailed by verified premises. (Ref.2 [24])
The principle states that any conclusion we make should maximize the relevant uncertainty
within constraints given by the verified premises. In other words, the principle guides us to utilize
all the available information but at the same time fully recognize our ignorance. This principle is
useful, for example, when we need to reconstruct an overall system from the knowledge of some
subsystems. The principle of maximum uncertainty is well developed and broadly utilized within
classical information theory,(Ref.2 [25]) where it is called the principle of maximum
entropy.(Ref.2 [9])
The last principle, the principle of uncertainty invariance, is of relatively recent origin (Ref.2
[21]). Its purpose is to guide meaningful transformations between various theories of uncertainty.
The principle postulates that the amount of uncertainty should be preserved in each
transformation of uncertainty from one mathematical framework to another. The principle was
first studied in the context of probability-possibility transformations (Ref.2 [20]). Unfortunately,
at the time, no well justified measure of uncertainty was available. Due to the unique connection
between uncertainty and information, the principle of uncertainty invariance can also be
conceived as a principle of information invariance or information preservation.(Ref.2 [22]).
1.4.
FUZZY MEASURE:
In mathematics, fuzzy measure considers a number of special classes of measures, each of which
is characterized by a special property. Some of the measures used in this theory are plausibility
and belief measures, fuzzy set membership function and the classical probability measures. In the
fuzzy measure theory, the conditions are precise, but the information about an element alone is
insufficient to determine which special classes of measure should be used (Ref.3[1]). The central
concept of fuzzy measure theory is the fuzzy measure which was introduced by Choquet in
1953(by Ref.3[1]) and independently defined by Sugeno in 1974 (by Ref.3[2]) in the context of
fuzzy integrals.
Consider, however, the jury members for a criminal trial who are uncertain about the guilt or
innocence of the defendant. The uncertainty in this situation seems to be of a different type; the
set of people who are guilty of the crime and the set of innocent people are assumed to have very
16
distinct boundaries. The concern, therefore, is not with the degree to which the defendant is
guilty, but with the degree to which the evidence proves his membership in either the crisp set of
guilty people or the crisp set of innocent people. We assume that perfect evidence would point to
full membership in one and only one of these sets. However, our evidence is rarely, if ever,
perfect, and some uncertainty usually prevails. In order to represent this type of uncertainty, we
could assign a value to each possible crisp set to which the element in question might belong.
This value would indicate the degree of evidence or certainty of the elements membership in the
set. Such a representation of uncertainty is known as a fuzzy measure.
In classical, two-valued logic, we would have to distinguish cold from not cold by fixing a strict
changeover point. We might decide that anything below 8 degrees Celsius is cold, and anything
else is not cold. This can be rather arbitrary. Fuzzy logic lets us avoid having to choose an
arbitrary changeover point essentially by allowing a whole spectrum of degrees of coldness. A
set of temperature like hot or cold is represented by a function. Given a temperature, the function
will return a number representing the degree of membership of that set. This number is called a
fuzzy measures.
For example: Some fuzzy measure for the set cold:
Temperature in 0C
-273
-40
0
5
10
15
100
1000
Fuzzy Measure
1
1
0.9
0.7
0.3
0.1
0
0
Cold
Cold
Not quite cold
On the cold side
A bit cold
Barely cold
Not cold
Not cold
The basis of the idea of coldness may be how people use the word cold perhaps. 30% of
people think that 100C is cold (function value 0.3) and 90% think that 0 0C is cold (function value
0.9). Also it may depend on the context. In terms of the weather, cold means one thing. In terms
of the temperature of the coolant in a nuclear reactor cold may means something else, so we
would need to have a cold function appropriate to our context.
17
Definition:
Given a universal set X and a non empty family
is a function
[1]
Boundary requirements:
[2]
Monotonicity: If
[3]
i=1
if i=1 Ai , then lim g( Ai )=g ( Ai )
i
[4]
A 1 A 2 ,
i=1
if i=1 Ai , then lim g ( A i )=g ( A i )
i
The boundary requirements [1] state that the element in question definitely does not belong to
the empty set and definitely does belong to the universal set. The empty set does not contain any
element hence it cannot contain the element of our interest, either; the universal set contains all
elements under consideration in each particular context; therefore it must contain our element as
well.
Requirement [2] states that the evidence of the membership of an element in a set must be at
least as great as the evidence that the element belongs to any subset of that set. Indeed with some
degree of certainty that the element belongs to a set, then our degree of certainty that is belongs to
a larger set containing the former set can be greater or equal, but it cannot be smaller.
Requirements [3] and [4] are clearly applicable only to an infinite universal set. They can
therefore be disregarded when the universal set is finite. Fuzzy measures are usually defined on
18
families
cases,
Three additional remarks regarding this definition are needed. First, functions that satisfy [1],
[2], and either [3] or [4] are equally important in fuzzy measure theory as functions that
satisfy all four requirements. These functions are called semicontinuous fuzzy measure; they are
either continuous from below or continuous from above. Second, it is sometimes needed to
generalize fuzzy measures by extending the range of function from [0, 1] to the set of
nonnegative real numbers and by excluding the second boundary requirement,
g ( X )=1 . These
generalizations are not desirable for our purpose. Third, fuzzy measures are generalizations of
probability measures or generalizations of classical measures. The generalization is obtained by
replacing the additivity requirement with the weaker requirements of monotonicity and continuity
or, at least, semicontinuity (Ref.3[5]).
A ,BX
, a fuzzy measure is
1.
Additive:
if
2.
super additive: if
g ( A B ) + g ( A B ) g ( A ) + g( B) for all A B= ;
3.
sub additive:
if
g ( A B ) + g ( A B ) g ( A ) + g(B)for all A B= ;
4.
super modular:
if
g ( A B ) + g ( A B ) g ( A ) + g ( B ) ;
5.
sub modular:
if
g ( A B ) + g ( A B ) g ( A ) + g(B) ;
6.
Symmetric:
7.
Boolean:
If | A|=|B| g ( A )=g ( B ) ;
If g ( A )=0g ( A )=1.
19
measures that every fuzzy measures g satisfies the inequality for any three sets
A , B , A B ,
g( A B) min [g ( A ) , g ( B ) ]
Similarly
A A BB A B
(1.15)
implies that every fuzzy measure g satisfies the inequality for any three sets
g( A B) max [ g ( A ) , g ( B ) ]
A ,B, A B ,
(1.16)
Understanding the properties of fuzzy measures is useful in application. When a fuzzy measure is
used to define a function such as the Sugeno integral or Choquet integral, these properties will be
crucial in understanding the function's behavior. For instance, the Choquet integral with respect
to an additive fuzzy measure reduces to the Lebesgue integral (Ref.3[4]). In discrete cases, a
symmetric fuzzy measure will result in the ordered weighted averaging (OWA) operator.
Submodular fuzzy measures result in convex functions, while supermodular fuzzy measures
result in concave functions when used to define a Choquet integral (Ref.3[2]).
Since fuzzy measures are defined on the power set (or, more formally, on the
algebra associated with X), even in discrete cases the number of variables can be quite high
(2 X ) . For this reason, in the context of multi-criteria decision analysis and other disciplines,
simplification assumptions on the fuzzy measure have been introduced so that it is less
computationally expensive to determine and use. For instance, when it is assumed the fuzzy
measure is additive, it will hold that
g ( E )= g( {i } )(1.17)
i E
20
and the values of the fuzzy measure can be evaluated from the values on X. Similarly,
a symmetric fuzzy measure is defined uniquely by
measures that can be used are the Sugeno- -fuzzy measure and
-additive measures,
If a problem can be solved in more than one way, all ways must lead to the same answer.
2.
The question posed and the way the answer is found must be totally transparent. There
must be no laps of faith required to understand how the answer followed from the given
informations.
3.
The methods of solution must not be ad-hoc. They must be general and admit to being
used for any problem, not just a limited class of problems. Moreover the applications
must be totally honest. For example, it is not proper for someone to present incomplete
information, develop an answer and then prove that the answer is incorrect because in the
light of additional information a different answer is obtained.
4.
The process should not introduce information that is not present in the original statement
of the problem.
21
The Axioms of Probability: The language systems, contains a set of axioms that are used
to constrain the probabilities assigned to events (Ref.3[12]). Four axioms of probability are as
follows:
1.
2.
Probabilities of an event that are necessarily true have a value of one, and those that are
necessarily false have a value of zero i.e. P(True) = 1 and P(False) = 0.
3.
4.
A ,B
A , B s .t . A B= .
The Joint Probability Distribution: The joint probability distribution is a function that
specifies a probability of certain state of the domain given the states of each variable in the
domain. Suppose that our domain consists of three random variables or atomic events
22
where
a1=1, a 2=0,a3=1
would be
a1 , a2 ,a 3
the many distributions that can be calculated using a probabilistic reasoning system.[Ref.2(20)].
(III)
Conditional Probability and Bayes Rule: Suppose a rational agent begins to perceive
data from its world, it stores this data as evidence. This evidence is used to calculate a posterior
or conditional probability which will be more accurate than the probability of an atomic event
without this evidence, known as a prior or unconditional probability.[Ref.3(12)].
Example:
If the subject does not have TB then the specificity of the test is 0.99.
From a large population, in which 2 in every 10,000 people have TB, a person is selected at
random and given the test, which comes back positive. What is the probability that the person
actually has TB?
Lets define event A as the person has TB and event B as the person tests positive for TB. It
P ( A )=
2
=0.0002P ( A )=1P ( A )=0.9998
.
10,000
The conditional probability P ( B| A ) ,the probability that the person will test positive for TB
given that the person has TB. This was given as 0.98. The other value we need
)
P ( B| A
, the
probability that the person will test positive for TB given that the person does not have TB. Since
23
a person who does not have TB will test negative 99% (given) of the time, he/she will test
positive 1% of the time and therefore
P ( A|B )=
)=0.01
P ( B| A
. By Bayes Rule as:
P ( A ) P ( B| A )
0.0002 0.98
=
) P ( B| A
) ( 0.0002 0.98 ) + ( 0.9998 0.01 )
P ( A ) P ( B| A )+ P ( A
0.0192 ( ) =1.92 ()
We might find this hard to believe, that fewer than 2% of people who test positive for TB using
this test actually have the disease. Ever though the sensitivity and specificity of this test are both
high, the extremely low incidence of TB in the population has a tremendous effect on the tests
positive predictive value, the population of people who test positive that actually have the
disease. To see this, we might try answering the same question assuming that the incidence of TB
in the population is 2 in 100 instead of 2 in 10,000.
(IV)
Conditional Independence: When the result of one atomic event, A, does not affect the
result of another atomic event, B, those two atomic events are known to be independent of each
other and this helps to resolve the uncertainty[Ref.3(13)]. This relationship has the following
mathematical property:
P ( A|B )=P ( A ) .
relationship.
Disadvantages with Probabilistic Method:
Probabilities must be assigned even if no information is available and assigns an equal amount of
probability to all such items. Probabilities require the consideration of all available evidence, not
only from the rules currently under consideration [Ref.2(20)]. Probabilistic methods always
require prior probabilities which are very hard to found out apriority [Ref.3(16)]. Probability may
be inappropriate where as the future is not always similar to the past. In probabilistic method
24
independence of evidences assumption often not valid and complex statements with conditional
dependencies cannot be decomposed into independent parts. In this method relationship between
hypothesis and evidence is reduced to a number [Ref.3(13)]. Probability theory is an ideal tool for
formalizing uncertainty in situations where class frequencies are known or where evidence is
based on outcomes of a sufficiently long series of independent random experiments [Ref.3(12)].
1.4.3. Evidence Theory:
Dempster-Shafer theory (DST) is a mathematical theory of evidence. The seminal work on the
subject is done by Shafer [Ref.3(8)], which is an expansion of previous work done by Dempster
[Ref.3(6)]. In a finite discrete space, DST can be interpreted as a generalization of probability
theory where probabilities are assigned to sets as opposed to mutually exclusive singletons. In
traditional probability theory, evidence is associated with only one possible event. In DST,
evidence can be associated with multiple possible events, i.e. sets of events. DST allows the
direct representation of uncertainty. There are three important functions in DST by[Ref.3(6)
&(8)]: (1) The basic probability assignment function (bpa or m), (2) The Belief function (Bel)
and (3) The Plausibility function (Pl). The theory of evidence is based on two dual non-additive
measures: (2) and (3).
(1) Basic Probability Assignment: Basic probability assignment does not refer to probability in
the classical sense. The basic probability assignment, represented by m, defines a mapping of the
power set to the interval between 0 and 1, s.t.
p: ( X ) [0,1]
properties:[Ref.2(20)]
(a)
( b)
p ( ) =0 , and
p ( A)=1
A ( X )
The value of p(A) pertains only to the set A and makes no additional claim about any subset of A.
Any further evidence on the subset of A would be represented by another basic probability
25
assignment. The summation of the basic probability assignment of all the subsets of the power set
is 1. As such, the basic probability assignment cannot be equated with a classical probability in
general [Ref.3(13)].
(2) Belief Measure: Given a measurable space (X , ) , a belief measure is a function
Bel : ( X ) [0,1] satisfying the following properties:
(a) Bel ( )=0 ,
i<k
(1.18)
Due to the inequality (1.18), belief measures are called superadditive. When X is infinite,
function Bel is also required to be continuous from above. For each
Y ( X ) , Bel(Y )
is
defined as the degree of belief, which is based on available evidence [Ref.3(6)], that a given
element of X belongs to the set Y. The inequality (1.18) implies the monotonicity requirement
[2] of fuzzy measure.
Let
now
X 1 X 3 for n=2
to (1.18), we get
26
Since
X 1 X 3= ,
Applying
Let
X 1= AX 2 = A
) Bel ( A A ) Bel( X) 1
Bel ( A ) + Bel ( A
)1
Bel ( A ) + Bel ( A
(1.19)
function
satisfying
the
following
properties:
(a)
Pl ( )=0 ,
(b)
Pl ( X ) =1 , and
( c ) Pl ( X 1 X 2 X n ) Pl ( X i ) Pl ( X i X k ) ++ (1 )n+1 Pl ( X 1 X 2 X n )
i
i<k
(1.20)
Due to the inequality (1.19), plausibility measures are called subadditive. When X is infinite,
function Pl is also required to be continuous from below [Ref.3(8)]. Let
(1.20) for n=2. Then we have,
)Pl ( A A
) Pl ( A A )
Pl ( A ) + Pl ( A
27
X 1= AX 2 = A
in
)Pl ( X ) Pl ()
Pl ( A ) + Pl ( A
) 1
Pl ( A ) +Pl ( A
(1.21)
According to inequality (1.19) and (1.21) we say that each belief measure, Bel, is a plausibility
measure, Pl, i.e. the relation between belief measure and plausibility measure is defined by the
equation[Ref.3(7)]
)
Pl ( A )=1Bel ( A
(1.22)
)
Bel ( A )=1Pl ( A
(1.23)
28
actual state of affairs) distinguishing what is plausible from what is less plausible, what is the
normal course of things from what is not, what is surprising from what is expected (Ref.3[15]). It
represents a flexible restriction on what is the actual state with the following conventions (similar
to probability, but opposite of Shackle's potential surprise scale):
(1) (s) = 0 means that state s is rejected as impossible;
(2) (s) = 1 means that state s is totally possible (= plausible).
For example, imprecise information such as Xs height is above 170cm implies that any height
h above 170 is possible any height equal to or below 170 is impossible for him. This can be
represented by a possibility measure defined on the height domain whose value is 0 if
h<170
and 1 if
h 170
in X is tall, the possibility can be accommodate degrees, the largest the degree, the largest the
possibility.
For consonant body of evidence, the belief measure becomes necessity measure ()
plausibility measure becomes possibility measure ( ) (Ref.3[19]).
Hence
Becomes
( A B )=min [ ( A ) , ( B ) ]
And
Pl ( A B )=max [ Pl ( A ) , Pl ( B ) ]
Becomes
( A B )=max [ ( A ) , ( B )]
And also
( A )=1 ( A ) ; ( A )=1 ( A )
Some other important measures on fuzzy sets and relations are defined here.
29
and
1.4.2.
Additive Measure:
be a measurable space. A function
Let X,
m:
m ( ) =0
If
An ,
then
n=1
m ( A n ) = m( An )
n=1
An
. A well-known example of
-additive
p ( X )=1 p ( A ) =1 p( A ' )
measures defined that are an important base of the XX century mathematics (Ref.3[9]). The
Lebesgue measures generalise the concept of length of a segment, and verify that if
n
Other measures given by Lebesgue are the exterior Lebesgue measures and interior Lebesgue
measures. A set A is Lebesgue measurable when both interior and exterior Lebesgue measures are
the same (Ref.3[11]). Some examples of Lebesgue measurable sets are the compact sets, the
empty set and the real numbers set R.
30
1.4.3.
-additive 1 X
for any
if its Mbius
M (F) 0 .
The -additive fuzzy measure limits the interaction between the subsets
EX
to
size E . This drastically reduces the number of variables needed to define the fuzzy
measure, and as can be anything from 1 to X , it allows for a compromise between
modelling ability and simplicity (Ref.3[3]).
1.4.4.
Normal Measure:
Let X,
A0
A
m( 0)=0.
1.
Am
m: [ 0,1 ]
in
and
A m =1
such that:
m ( A m )=1.
2.
by (Ref.3[14]):
31
g : [ 0,1 ]
verifying
g ( )=0g ( X )=1
1.
2.
If
3.
If
An
and
A 1 A 2
then
Property 3 is called Sugenos convergence. The Sugeno measures are monotone but its main
characteristic is that additivity is not needed. Several measures on finite algebras, as probability,
credibility measures and plausibility measures are Sugeno measures. The possibility measure on
possibility distributions introduced by Zadeh (Ref.3[7]) gives Sugeno measures.
1.4.6.
Fuzzy Sugeno
-Measure:
1.5.
g : [ 0,1 ]
g ( A B ) =g ( A )+ g ( B ) + g ( A ) g (B)
is a fuzzy
and let X ,
The concept of fuzzy measure does not require additivity , but it requires monotonicity related to
the inclusion of sets. Additivity can be very effective and convenient in some applications, but
can also be somewhat inadequate in many reasoning environments of the real world as in
approximate reasoning, fuzzy logic, artificial intelligence, game theory, decision making,
psychology, economy, data mining, etc., that require the definition of non-additive measures and
32
large amount of open problems. For example, the efficiency of a set of workers is being
measured; the efficiency of the some people doing teamwork is not the addition of the efficiency
of each individual working on their own.
The theory of non-additive set functions had influences on many parts of mathematics and
different areas of science and technology. Recently, many authors have investigated different type
of non-additive set functions, as sub-measures, k-triangular set functions,
- decomposable
measures, null-additive set functions and others. The range of the null-additive set functions,
introduced by (Ref.4[6]). Suzuki introduced and investigated atoms of fuzzy measures (Ref.4[4])
and Pap introduced and discussed atoms of null-additive set functions (Ref.4[3]).
Definition:
Let X be a set and a
2.
A B= , is always null-additive.
A generalization of -decomposable measure from example 1 is the decomposable measure m: [a , b ] . Namely, let [a, b] be a closed real interval,
33
It is
A , B , and
A B= . Then m is null-
additive.
3.
Let m ( A ) 0 whenever
4.
Let
A , A . Then m is null-additive.
Saks decomposition:
1.
then
m ( B )=0 ,or
2.
m ( B )=m( A) , what is
For
A ,
>0
and
m ( A ) m ( A B ) m ( A )+
(resp.
m ( A ) m
holds (Ref.4[5])
Remark:
If m is a fuzzy measure, then we can omit in the preceding definition the supposition
A B= (resp. B A ).
The set function m is a null-additive non fuzzy measure which is autocontinuous from above and
from below. The autocontinuity from above (below) in the sense of Wang (Ref.4[5]), if
A {B
m ( A Bn ) m ( A ) (resp . m ( n ) m ( A ))
whenever
If m is a finite fuzzy measure, then m is autocontinuous from above (below) iff m is Wautocontinuous from above (below). If m is a finite fuzzy measure, then m is autocontinuous
from below iff it is autocontinuous from below (Ref.4[5]).
Each set function which is autocontinuous from above is null-additive but there exists nulladditive fuzzy measure which is not autocontinuous from above (Example 1). If the set X is
countable, then null-additivity of a finite fuzzy measure is equivalent to the autocontinuity
(Ref.4[6]).
35