Fuzzy Set Theory in Database

Systems
Seminar Presentation
MA7355 Fuzzy Set Theory and Applications
Name : Fatima Basheer
Roll No : B160262CS
Motivation
● To handle information that is less than ideal - incomplete, indeterministic,
contradictory, vague, imprecise, etc.
● Databases to store and manipulate not only precise facts but also subjective
expert opinions, judgements, and values.
● To use database as a decision aid in areas such as medical diagnosis,
employment, investment and geological exploration where such “soft” and
subjective data are valuable.
● Relieve the user of having to formulate precise queries.
Database systems incorporating imprecision
should be able to propagate the level of uncertainty
associated with data to the answers or conclusions
based on the data.

Precise answers should not be generated

from imprecise data!!!
Classical Relational Database
Columns correspond
to fields, attributes
Other Relational called domains
Table : Employees
Operators
Complement, Union,
Intersection and Join

Rows correspond to
records or entries
called tuples

Project (EMPLOYEES : NAME, FNAME, SALARY) where CITY = “3”

Obtain name, fname and salary of all employees whose city is 3 by constructing a new relation with
domains name, fname and salary as a projection of the original relation.
Fuzzy Relational Database Model
● Proposed by Buckles and Petry in late 20th century.

● Differs from crisp model in 2 ways :

○ Elements of tuples contained in the relations maybe subsets of the domain universal set.

○ A similarity relation is defined on each domain universal set.

I Qualification :
Elements of the tuples contained in the relations may be subsets of the domain
universal set.

This means the elements of the tuples could be singletons (classical) or crisp subsets of the domain universal
sets as in MARKETS with the domains AREA, SIZE, POTENTIAL represented in the following table.

Domain Values not singletons may indicate merging of opinions or judgements.

RELATION : MARKETS
AREA SIZE POTENTIAL

East Large Good

Midwest {Large, Medium} {Moderate, Good}

South Small {Good, Excellent}

II Qualification :
A similarity relation is defined on each domain universal set.

In classical DB model, crisp equivalence relation is defined on each domain universal set which groups
elements which are strictly equivalent. This is used to eliminate or ignore redundant tuples.

Most often the equivalence classes are the singleton sets of universal set itself. In fuzzy DB model, it is
generalized to a fuzzy similarity relation.

Along with the 4 components of conventional relational algebra, the new operator specifies for a threshold
level defining the minimum acceptable degree of similarity between elements in some specified domain.
Example
Our database contains the opinions of a group of experts on three policy options X, Y and Z. Two relations
are contained within the database: EXPERT, which has domains NAME and FIELD and which associates
the name and field of each expert, and ASSESSMENT, which has domains OPTION, NAME, and OPINION
and associates the name of each expert with their expressed opinions on the policy options. The similarity
relation for the domain OPINION on the domain universal set {highly favorable (HF), favorable (F), slightly
favorable (SF), slightly negative (SN), negative (N), and highly negative (HN)} is given.

HF F SF SN N HN

HF 1 .8 .6 .2 0 0
F
SF
.8 1 .8 .6 .2 0 QUERY
.6 .8 1 .8 .6 .2
SN
N
0 .2 .6 .8 1 .8 Which sociologists are in
HN 0 0 .2 .6 .8 1 considerable agreement with Kass
concerning policy option Y?
Relations in Relational Database
Relation : ASSESSMENT
OPTION NAME OPINION OPTION NAME OPINION
Relation : EXPERT
X Osborn Favorable Y Feldman Slightly negative
NAME FIELD
X Fee Negative Z Osborn Negative
Cohen Sociologist
X Fadem Slightly favorable Z Kass Slightly negative
Fadem Economist
X Feldman Highly favorable Z Fee Slightly favorable
Fee Attorney
Y Cohen Slightly negative
Feldman Economist
Y Osborn Slightly favorable

Kass Physician
Y Fee Highly favorable

Osborn Sociologist Y Schreiber Favorable

Schreiber Sociologist Y Kass Favorable

Specterman Sociologist Y Fadem Negative

Y Specterman Highly favorable

Step 1:
Retrieve the opinion of Kass concerning option Y.

R1 = (Project (Select ASSESSMENT where Relation : R1 OPINION

NAME = ‘Kass’ and OPTION = Y over OPINION) Favorable

Step 2: Relation : R2 NAME

Selection of all sociologists from the table of experts. Osborn

Schreiber
R2 = (Project (Select EXPERT where FIELD =
‘Sociologist;) over NAME) Cohen

Specterman
Step 3: NAME OPINION

Construct a temporary relation R3 on domains NAME Osborn Slightly

favorable
and OPINION, which lists the opinions of the
Schreiber Favorable
sociologists in R2 about option Y.
Cohen Slightly
(Project (Select (Join R2 and ASSESSMENT over negative
NAME) where OPTION = ‘Y’) over NAME, Specterman Highly favorable
OPINION)
Step 4:
Join of relations R1(giving opinion of Kass) and
NAME OPINION
R3 (giving opinion of the sociologists) that
specifies a threshold similarity level of .75 on the
domain OPINION, which is chosen in this example
{Osborn, {Slightly
to represent condition of considerable agreement. Schreiber, favorable,
Specterman} favorable, highly
(Join R3 and R1 over OPINION) with favorable}

THRES(OPINION) >=.75 and

THRES(NAME) >= 0

The specification of a zero similarity threshold

level for NAME is to allow merging of names into
a set.
Note that the result is less precise and contains less information than a response of

NAME OPINION

Osborn Slightly favorable

Schreiber Favorable

Specterman Highly favorable

In this way, the uncertainty contained in the specification of considerable agreement

and in the similarity defined over the possible opinions is propagated to the response
given.

It is an area of extensive research and a lot more is going on!!!

References
❏ Fuzzy Sets, Uncertainty and Information, authored by Geroge J Klir, Tina A
Floger