SOFT COMPUTING

MODULE 4
FUZZY Membership
Functions
Topics
Fuzzy membership functions
Features of membership function
Fuzzification
Methods of membership value assignments
λ- Cut or α- Cut
Defuzzification
Membership functions

Membership function defines the fuzziness in a

fuzzy set irrespective of the elements in the set,
which are discrete or continuous.
They are generally represented in graphical form.
The rules that describe fuzziness graphically are also
fuzzy.
 Fuzzy membership
functions
Membership function on a discrete universe of course is trivial.
However, a membership function on a continuous universe of
discourse needs a special attention.
Following figures shows a typical examples of membership functions.

µ
µ

x x x

< triangular > < trapezoidal > < curve >

µ

x x

< non-uniform > < non-uniform >

Soft Computing Applications

Features of membership funcion
Core
Support
Non convex
Linguistic variables and hedges
Linguistic Variable
A linguistic variable is
one with a value that is a
natural language
expression referring to
some quantity of interest.
So the value of the
linguistic variable is a
word or a sentence. This
is the main difference
between a linguistic
variable and a numerical
one.
Linguistic variables and hedges
Linguistic Variable
 Generally, a linguistic variable is a composite term
u = u1, u2, ....,un which is a concatenation of atomic
terms u1, u2, ...,un
 atomic terms can be divided into four categories:
o primary terms, which are the labels of specified
fuzzy subsets of the universe of discourse (e.g. small
and big);
oConnectives: AND, OR and the negation NOT;
o hedges: such as VERY, MOST, RATHER,
SLIGHTLY, MORE OR LESS, etc.;
o markers such as parenthesis.
Linguistic variables and hedges
primary terms
A primary term is actually just a name or a label of a
fuzzy set. It usually describes the word which is
used by experts to express their opinion about the
value of one of the object characteristics, e.g : old,
large, fast, etc.

20
Linguistic variables and hedges
connectives
The connectives (AND, OR, and NOT) realise the
operations of intersections, union, complement
considered earlier.
 Example :How can we express
 slow or medium ?

not fast and medium ?

Linguistic variables and hedges
hedges( modifiers)
Hedges are used to produce a larger set of
values for a linguistic variable from a small
collection of primary terms through the
processes of intensification or concentration,
dilation and fuzzification.
For example, the operator ‘very’ is usually
defined as a concentration operator as:not always
This is one way

very u = u2
This operator can also be composed with
itself, thus:
very (very u) = (very u) 2 = u 4
Linguistic variables and hedges
hedges (modifiers)
Hedges (example): The composite term ‘very old’ can

be obtained from the term old as
very old = old2
Example : consider a fuzzy set

young   020.8 0 .6
30
0.2
40
0
50


very young   020.64 0.36

30
0.04
40
0
50


very very young   0.4096

20
0.1296
30
0.0016
40
0
50

Another example

Extremely A = A3
Linguistic variables and hedges
Wind is a little strong.
Weather is quite cold.
Height is almost tall.
Weight is very high.
Wind, Weather, Height and Weight are
linguistic variables.
A little, Quite, Almost, Very are hedges.
Strong, Cold, Tall and high are linguistic
value.
Example
Linguistic Linguistic Linguistic Linguistic
Variable Value Variable Value

cold cheap
if temperature is cold and oil is cheap

high
then heating is high
Linguistic Linguistic
Variable Value
Definition [Zadeh 1973]
A linguistic variable is characterized by a quintuple

 x, T ( x),U , G, M 
Name

Term Set
Universe
Syntactic Rule
Semantic Rule
 Example
A linguistic variable is characterized by a quintuple

 x, T ( x),U , G, M 
age
Example semantic rule:
old, very old, not so old, 
 
G (age) more or less young,  M (old)   u, old (u )  u  [0,100]
quite young, very young 
 
 0 u  [0,50]
 1
old (u )    u  50   2 
[0, 100]  1   5   u  [50,100]
    
Example
Membership of body fitness
Methods of membership value assignments

1 Intuition
2 Inference
3 Rank ordering
4 Angular fuzzy
sets
5 Neural networks
6 Genetic
algorithm
7 Inductive
reasoning
Intuition

Intuition method is based upon the common

intelligence of humans.
It is the capacity of the human to develop membership
func- tions on the basis of their own intelligence and
understanding capability.
There should an in depth knowledge of the application
to which membership value assignment has to be
made.
Membership functions for the fuzzy variable
"weight"
Problems

(1) Using your own intuition and definitions of

the uni- verse of discourse, plot fuzzy
membership functions for " weight of people " .
 (2) Using your own intuition and definitions of
the uni- verse of discourse, plot fuzzy
membership functions for " age of people " .
Inference

The inference method uses knowledge to

perform deductive reasoning.
Deduction achieves conclusion by means of
forward inference.
The knowledge of geometrical shapes and
geometry is used for defining membership
values.
The membership functions may be defined
using various shapes: triangular, trapezoidal,
bell shaped,Gaussian etc.
The inference method here is via triangular
Problem
Rank Ordering

On the basis of the preferences made by an

individual, a committee, a poll and other
opinion methods.
Pairwise comparisons enables to determine
preferences.
 This results in determining the order of the
membership.
Eg
1. Formation of Government is based on polling
concept
2. To identify a best student, ranking may be
performed
3 To buy a car we can seek several opinion
Angular Fuzzy Sets
Consider pH value of wastewater from a dyeing
industry.
pH = 7 → neutral; below 7 → Acid; 7-14 →Base
 Linguistic variables are build in such a way that
neutral (N)→ ᶿ = -Π/2 rad

Base (B)
Very Base (VB)
Exact Base(EB)
Medium Base (MB)
Acid (A)
Very Acid (VA)
Exact Acid(EA)
Medium Acid (MA)
 Neural Networks
NN can also be used to obtain fuzzy
membership values
Fuzzy MFs are created for input dataset
I/p data set divided in to training and test
data
Data points are grouped in to clusters
If a data point belong to a cluster its MF is
1 in that cluster and 0 in other clusters
NN uses data point marked 1 for training
When a coordinate location of a point is
given, NN assigns a membership value to
that point
NN classifies the point into one of the
Genetic Algorithms
Induction Reasoning
Uses backward inference
Employs entropy minimization principle
Well defined database for i/p – o/p
relationship exists
Can be applied for complex systems where
data are abundant and static
Not suited for dynamic systems
There are 3 laws of induction
Membership function is calculated as
 Defuzzification

Defuzzification means the fuzzy to crisp conversion.

The fuzzy results can not be used in an application, where

decision has to be taken only on crisp values.
Example:

If THIGH then rotate RFIRST .

Here, may be input THIGH is fuzzy, but action rotate should be

based on the crisp value of RFIRST .
 Defuzzification
methods

A number of defuzzification methods are known. Such as

1 Lambda-cut method

2 Weighted average method

3 Maxima methods

4 Centroid methods

Soft Computing Applications

 method

•Lambda-cut method is applicable to derive crisp value.

• It can be applied to both fuzzy set and fuzzy relation

Lambda-cut method for fuzzy set Lambda-cut method for fuzzy

relation

Lambda-cut method is alternatively termed as Alpha-cut method.

 Lamda-cut method for fuzzy
set

1
In this method a fuzzy set A is transformed into a crisp set A λ for a
given value of λ (0 ≤ λ ≤ 1)

2
In other-words, A λ = {x|µA(x ) ≥ λ}

3
That is, the value of Lambda-cut set A λ is x , when the
membership value corresponding to x is greater than or equal to
the specified λ.
4
This Lambda-cut set A λ is also called alpha-cut set.

Soft Computing Applications

 Example

A1 = {(x1, 0.9), (x2, 0.5), (x3, 0.2), (x4, 0.3)}

Then A0.6 = {(x1, 1), (x2, 0), (x3, 0), (x4, 0)} = {x1}

and

A2 = {(x1, 0.1), (x2, 0.5), (x3, 0.8), (x4, 0.7)}

A0.2 = {(x1, 0), (x2, 1), (x3, 1), (x4, 1)} = {x2, x3, x4}

Soft Computing Applications

 Lambda-cut sets :
Example
Two fuzzy sets P and Q are defined on x as follows.

Find the following : ( Home Work )

(a) P0.2, Q0.3

(b) (P ∪ Q)0.6

(c) (P ∪ P)0.8

(d) (P ∩ Q)0.4

Soft Computing Applications

Lamda cut for a Fuzzy Relation
 Properties of λ- cut in fuzzy relation

Lambda-cut method converts a fuzzy set (or a fuzzy

relation) into crisp set (or relation).
 Defuzzification
methods

1 Lambda-cut method

2 Weighted average method

3 Maxima methods

4 Centroid methods

Soft Computing Applications

Defuzzification Methods contd.

Maxima Methods
1 Height method
2 First of maxima (FoM)
3 Last of maxima (LoM)
4 Mean of maxima(MoM)

Centroid methods ƴ
1 Center of gravity method (CoG)
2 Center of sum method (CoS) ƴ
3 Center of area method (CoA)

Weighted average method

Soft Computing Applications

 Maxima methods

Height method
First of maxima (FoM)
Last of maxima (LoM)
Mean of maxima(MoM)

Soft Computing Applications

Maxima method : Height
method
This method is based on Max-membership principle, and defined
as follows.

µC (x∗ ) ≥ µC (x ) for all x ∈ X

c

Note:
1. Here, x ∗ is the height of the output fuzzy set C.
2. This method is applicable when height is unique.
Soft Computing Applications
Maxima method : FoM

FoM: First of Maxima : x ∗ = min{x|C(x ) = maxw C{w}}

c

Soft Computing Applications

Maxima method : LoM

LoM : Last of Maxima : x ∗ = max{x|C(x ) = maxw C{w}}

c

Soft Computing Applications

Maxima method : MoM
MoM : Example
1

Suppose, a fuzzy set Young is defined as follows:

Young = {(15,0.5), (20,0.8), (25,0.8), (30,0.5), (35,0.3) }

Then the crisp value of Young using MoM method is

2 = 22 .5
x ∗ = 20+25

Thus, a person of 22.5 years old is treated as young!

Soft Computing Applications

MoM : Example
2
What is the crisp value of the fuzzy set using MoM in the
following case?

c

x∗ = a+b
2

Note:
Thus, MoM is also synonymous to middle of maxima.
MoM is also general method of Height.
 Centroid methods

Center of gravity method (CoG)

4

Center of sum method (CoS)

 Center of area method (CoA)
2

Soft Computing Applications

 Centroid method :
CoG
1
The basic principle in CoG method is to find the point x ∗ where a
vertical line would slice the aggregate into two equal masses.
2
Mathematically, the CoG can be expressed as follows :

3 Graphically
,

Center of gravity
c

x*

x
Centroid method : CoG
contd.
CoG : A geometrical method of
calculation

Soft Computing Applications

calculation ( HW)
CoG: An example of integral method of
calculation ( HW)
CoG: An example of integral method of
calculation ( HW)
 CoS
If the output fuzzy set C = C1 ∪ C2 ∪ ....Cn, then the crisp value
according to CoS is defined as

c2
c3
c1
A2 A3
A1
5

x1 x2 x3

Soft Computing Applications

 Centroid method :
CoS

Note:

1 In CoG method, the overlapping area is counted once, whereas,

in CoS , the overlapping is counted twice or so.

2 In CoS, we use the center of area and hence, its name instead
of center of gravity as in CoG.

Soft Computing Applications

CoS:
Example

Soft Computing Applications

Centroid method: Certer of largest area
Weighted average method
Weighted average method

Graphically
,

Soft Computing Applications

Summary
CoG

 CoA


Weighted Average Method