Lecture 23

IIT Kanpur July 201
20 followers
Feb 16, 2024
[Music]
uh a very good morning good afternoon
good evening to all the participants and
the students for this nptl course titled
multicriteria decision
making and my good name is raghunandan
Gupta from the II department at it kpur
so as you know this is the fifth week
set of classes which is going on and
this total course duration is for 12
weeks spread over 60
lectures and each week we have five
lectures of half an hour each and as you
know there are assignments after each
week so obviously once you finish the
fifth week you will all the students and
participants will take the fifth set of
assignments so the broad umbrella under
which we are discussing many of the
topics is basically multi criteria
decision making and under that you have
multiobjective decision making
multiattribute decision making and there
are subp Parts like of mcdm and I have
explained that that in multiobjective
decision making is more
mathematical where continuous variables
are considered more straightforward word
and it is more objective while
multiattribute decision making and the
concept related to multi attribute util
theory is more subjective more discret
search and there is much more the
answers which you may get under
madm and the concept used in
maut would be much more um subject based
on the decision maker this is the 23rd
lecture which is we are in the third
lecture for the fifth week the coverage
which we will be doing and obviously
when I mention the coverage I mention
the point also that overall under the
broad umbrella we will cover Merit
optimity we have covered few of the
concepts of mcdm definitions the
comparison of M M modm then
multi-attribute um decision making
and we'll consider further on the parit
optimality con concept property of
dominance what does dominance mean
strong parot optimality weak parot
optimality how the concepts of parot
optimality can be understood in 2D space
in 3D space uh it will be we can imagine
and we can also analyze but in higher
Dimension it will be difficult for us to
visualize we'll consider effective
versus in in in efficient Solutions the
concept of Kish condition this what I'm
saying is not the total coverage for
today's lecture we will cover it and
whatever today's lecture is that will be
updated in the slides and it will be
shared accordingly the karush kakar
condition the scales of measurements
which are nominal scale ordinal scale
interval scale ratio scale and why
scales are important you will understand
when we come to the concept of multi
multiple attribute concept or decision
making where decisions are more
subjective than being objective and then
we'll go into dep about goal programming
and go step by step so when we consider
a multiobjective problem which we did
discuss and I did mention few important
points which will be highlighted with
examples so the points which we
discussed and which was the focus I'll
use the red color so that the
differentiation of the color scheme
which I always insist would make things
much
clear so if you remember in multi
objective decision making where it is
more
objective uh oriented multiobjective
decision making
multi where it is more as I said
continuous mathematical much more clear
clear-cut
answers so you have
functions fub1 to say for
example f j j is a suffix I'm using even
though the concept where we used in the
last lecture was M but J is being done
SS that all the subscripts are
absolutely clearcut there's no
confusion and subject to conditions
being G1 being less than equal to B1 and
these uh continue to sa for example g k
capital K being less than equal to B
capital
K and if you remember I did also mention
X is not a scalar decision variables is
basically in the search space from X1 to
x capital L so it's a l Dimension one
but obviously remember when we solve the
problems all always later on in multi
attribute uh decision making we'll
always consider there are m capital M
number of
Alternatives which is
a and there are n number of
criteria and the relationship of M and N
which I said that M can be equal to n m
can be more than n m can be less than n
will be clarified as we proceed with
different type of problems
so let us consider the multiobjective
problem of the form where you have you
need to optimize optimize means it can
be maximization or minimization there's
no problem you need to optimize F1 to FM
which is basically F1 to FJ we are
saying X which is the decision variable
by the way X when we are considering it
can be depending on the problem can be
continuous can be discrete can be binary
can be mixed integer whatever it is so
when I consider these Pro problems I'm
mentioning the examples they may be
repetition plus please bear with me
continuous case can me say for example
you are producing pains in liters liters
can be 2.52 lit it can be 10.69 L it can
be
125.638 variables greater than zero now
when I come to the concept of integers
it is when when you are considering the
production of C for example number of
chairs number of tables number of trucks
being utilized when you're transporting
goods number of bins you're packing
packing that means putting in and
transporting those are integers that
means you cannot have 2.25 number of
trucks being being utilized for
transportation or you cannot basically
make 1.25 n number of chairs so they are
integers in the case of many problems
where is binary it will be whether you
build the factory not build the factory
whether you transport do not transport
so they are binary and in many of the
other examples they can be decision
variables which we'll see later on which
can be integers some of them integers
some of the discrete some of the binary
and and accordingly so when I mention X
is basically uh of L Dimension so I'm
not going to the details but this is the
basic and the framework here fub1 to FM
which is basically F1 to FJ are
individual objective functions such that
I want to find out the best value of
these functions which we considered in
the example the LA last lecture so if
you remember there were two objectives
and these two objectives when they're
optimized they give you different DEC
set of decision variables X1 and
X2 and uh interestingly the
maximization problem for both F1 and FS2
gave the same answer for F1 and F2 but
for different um solution or for the
decision space but when we try to
basically find out the common Point
based on the multiobjective case then we
saw that it was a integer problem if you
remember then we found that the
objective functions for both this uh F1
and FS2 were different but what was
important is the decision variable which
was X1 and X2 was same and I showed you
in a very simple way in twood
dimensional case how you can visualize
how uh the solution space that's Point
number one and also the optimum point or
Optimum solution separately for fub1
separately for FS2 and then combining
them we saw that there was a difference
so here we are saying that fi uh which
is basically each objective function
basically is on the real line once you
put the value of the decision variable
it gives you some functional value which
you want to optimize individually it can
be done but we want to basically
collectively optimize that's what the
idea of multicriteria or multiobjective
decision making is there may be
instances when the objective functions
are at least partially in conflict that
is we do not have any decision Vector
X which optimize all of the objectives
problems or all object objective
functions
simultaneously so again coming back to
these two examples number one the
example we which we consider in the 22nd
lecture where we want to optimize a
function F1 and FS2 separately first and
then combining them so we found out that
they were different results now when we
come to the concept of multiattribute
concept where there are attribute
characteristics and where it is much
more subjective consider these examples
that my my means an individual decision
maker and consider I am the decision
maker I want to buy a car or I want to
get an admission for some college or I
want to buy a house or I want to recruit
somebody so in that case if when I'm
considering buying a car the conflicting
um criteria here the the conflicts are
happening for the criteria or it can
happen for the Alternatives when we
basically combine the criterias and try
to basically compare the Alternatives
alternatives are basically buying the
car can be based on price can be based
on mileage can be based on maintenance
cost can be based on safety and so on
and so forth for the case of mileage
case of Maintenance case of cost it is
all very clearcut there's a price
component and it can be communicated it
is very objective in nature and you can
analyze but what if we consider the
concept of safety features what we
consider the concept of uh say for
example style what is the idea that how
would we consider the concept of color
so in that case the concepts of
multi-attribute decision making along
with multiobjective decision making will
be considered for such decisions when
the decision is there to buy a car the
other two examples which I mentioned few
minut uh minutes back is buying the flat
I come back to the same example time and
again because these are the example
based on which we are proceeding so it
should definitely give you a better idea
that how these problems can be utilized
or the concepts can be utilized to solve
the problem when you're solving the
problem of buying a flat on an apartment
we know the objective facts are price
objective facts are maintenance cost but
the subjective facts can be how safe it
is or how near it is to the Metro or how
near it is through the Transportation
Hub like bus taxi Auto or how close are
the schools how close is are the set of
offices where people work where you have
to basically go and work so these are uh
the sets of criterias we want to
basically consider when trying when you
want to buy an apartment similarly when
you are going for higher studies as I
mentioned I'll keep mentioning these
points again it can be based on price of
cost of tution it can be based on what
is the quality of Education what are the
different type of courses which are
offered how are the faculties what is
the scenario and the chance of getting
good internship what are the scenario
and chance of get a getting a good um uh
final placement what is the contact and
how good is the contact of the alumni so
these all these criterias will be
considered where some of the criterias
would be objective some of them would be
subjective but coming back to the last
point which is is there there may be
instances where there are conflicts for
the car example price is too high which
is detrimental to to my decision process
because I want to basically buy at a
lower price but the safety features are
excellent or consider there the safety
features are very low which is I do not
want but the mileage is excellent which
is positive to me so how you basically
balance the positive and negative points
that has to be analyzed for buying the
for getting an admission in a college
it's very costly but the set of of
alumni or set of of courses or the set
of Faculty members are excellent or it
may be the cost of getting a loan is
very cheap because people are willing to
banks are willing to give you very good
um um rate of interest rate for getting
good um loans but maybe the educational
level or the type of education which
you'll get in that Institute maybe of
not not that quality when you basically
try to analyze with other institutes so
those compromises have to be analyzed
and what is best that would we will
consider considering the concepts of
utility Theory which have already
considered and further on to make things
more formal we State few results which
are basically the properties of parit
Optimal Solutions which may be
considered as relevant and important in
this area now when I'm considering the
concept of parent optimal Solutions if
you remember generally consider this
case and I'll draw the diagram using a
different color let me draw the
axis
here so there are two
products and then mention them as
product those p p is not the price it's
product just what one product one and
product
two now consider your budget is
fixed and you want to uh spend money for
your consumption and as savings so
obviously if your budget your total
income your total salary is fixed and
obviously if your expenditures increase
then your savings decrease and if your
and consider if your expenditure
decreases obviously you will try to
increase your savings consider another
example you want to consume some uh
amount of hot beverage so this is a very
common example say for example t or
coffee if the prices of tea increases
obviously switch to coffee and vice
versa depending on that so if you
consider the concept of par Optimal
Solutions in the two two dimensional
space and these Concepts were definitely
highlighted in few of the examples which
we did in the first set of lectures in
the first week so generally depending on
your indifference
curve or your level of budget we will
basically have i1 which is indifference
one then I2 then I3 and so on and so
forth where as you move move from i1 to
I2 I2 to I3 obviously it means your
level of satisfaction is increasing but
the balance which you want to do between
product one product two tea coffee or
expenditure and
consumptions
um expenditures and savings should be
balanced in such a way that you maintain
that parity
now these curves can be extended for the
higher dimensional wall so say for
example in the three dimension it will
look like a circus tent which is just an
envelope and any point there which you
consider would be the optimum Point
based on what is your level of
satisfaction which you are trying to
meet based on all the three products
because in that case there would be
product one product two product three in
this example there are only two products
it is in a two dimensional cartisian
coordinate now continuing these examples
so the three properties which we will
mention here theoretically and later on
it will be highlighted as we consider
the examples by the way when I mention
the concept of highlighting it will come
up naturally as we solve the problems
the first property is property of
dominance so we will say in mathematical
terms a vector X or the the decision
which you're going to take will is said
to
dominate a vector y so X and Y are two
sets such
that that the functional values of FX
and
FY in in maximum of the cases maximum
when I'm mentioning I'll come to that
later fi or and or fiix or fi y so f I
is basically any objective function
which you have you have basically J
number of objective functions we are
collect considering each of them
individually so it will give you the
fact that for any I from I is equal to 1
to J or of any J small J is equal to 1 1
to capital j f j x will dominate or will
be greater than in value then FJ y I'm
calling it J because we are considering
the number of objective function as J
small J is equal to 1 to Capital J here
it is basically let me Mark it with a
different
color this is J small J this is 1 2
capital
J so it will be greater than equal to
sign and it will be strictly greater for
at least one of these Capital
JS which means
that in case if you have the dominance
the functional value based on X when you
compare with Y can be equal for all the
cases except one in that case we'll say
the set solution of x will dominate y
now if I consider the concept of strong
so property of dominance basically if I
go into further on in the concept of
strong parro
optimality we'll consider a vector X
again a
solution xar X xar is basically a
vector which is element of x x is the
set of solutions which you have is
defined to be as strong of parit optimal
the word strong is important if there
exists some extra such that it dominates
all the other x's in capital
x and this concept of strong optimality
point would basically prove that there
there are decision variables X star
which would be denoted as subsets of
X so this concept of subsets would be
used in such a way that they can be
proper subsets or just common subsets
which are there now the concept of weak
par optimality this ideas which I'm
explaining would be made clear with the
examples which I said a vector xstar in
X is defined as weak par optimality let
me use a different color say for example
blue if there exist no other Vector of X
such that f iix is less than not less
than equal to is less than x
star and this objective Vector would be
called a weak parit
optimality if it corresponds to X star
and is weak par optimality based on the
fact it is less in the other case it was
basically based on the fact
that that less than sign can be replaced
by less than equal to sign also now when
we consider the concept of par
optimality we will always consider very
simply the set which would dominate or a
set of solutions which will dominate the
other Solutions in at least one of these
values of of X based on the the concept
of different type of objective functions
which I have so say for example if I
have very simply set one which I'm
denoting in black color and the
objective values are given here so I'm
not considering the corresponding X for
the time
being so fub1 is
to based on that particular X
star one so this is xar one star I'm
just utilizing in order to denote the
second value based on X1 star and
remember X1 star which I'm writing is
basically a
vector so the second objective function
is say for example three and the third
objective function is basically
four when I want to compare with a
different set of solution and I'll use a
different color consider it is X2
star and X1 X2 are vector vectors again
they are a subset of the overall set of
solutions and the solutions of fub1 FS2
F3 are now
considered
2 3 and
three which would mean X1
star does not dominate X2 Star based on
functional value one because both are
two they are equal to X1 star does not
dominate X2 Star based on functional
value FS2 because both the values are
three but
interestingly which I'll Mark here with
a different
color the value of
F3 based on X1 star and when I compare
the value of F3 based on x2 star 4 is
greater than three so in that case we
can we would say that X1 star would
dominate the value
of X2 star but where the problem would
lie so here there's no confusion in the
sense that 2 is equal to two 3 is
basically equal to three I'm talking
about the values of FS1 F2 F3 and 4 is
greater than three so X1 star dominance
X2 star but now consider another example
so I'll use a different
color and let me Mark it in green and
marking on the top consider X3
star the values of fub1 FS2 F3
are when I'm going to compare with X1
star
only now here it is
interesting if I compare the values of
X1 star X3 star for F1 F2 F3 if you see
X1 star when I want to find out X F1 is
two but when I compare with the value of
FS1 based on X3 star it is three so
obviously X3 star which is noted in
green color is
dominating next when I come to FS2
functional value both the values for X1
star and X3 star are three so they are
equal when I come to the concept of the
function value of F3 based on X1 star
and X3 star you will find X1 star is
four while the value of x3 star is three
so the
dichotomy or the decision based on I'll
use a different color based on the fact
that three is greater than two here
which is fine but here four is greater
than three which basically tilts the
balance again in favor of x X1 star but
the question is that the net worth of
three being greater than two or four
being greater than three has to analyze
where the concept of parit optimality
and dominance has to be considered so I
may say I as an individual would say
that the functional value which I have
for fub1 three is better other person
may say no three is not better maybe
four is better than three when I'm
talking about four and three if you look
at this set of diagrams which I have
drawn the confusion should be not
there now we'll consider a very simple
two-dimensional figure to analyze the
ideas of parit
optimality so it will illustrate the par
optimality plots for four different
combinations and will denote the
objective functions now we are going to
consider the objective functions by
using objective 1 underscore objective
two underscore where objective one and
objective two are the two different sets
of values which I want so let us
consider objective function one and
objective function two in a very simple
sense this is
fub1 and this is FS2 so in the other
example in the other slides we consider
fub1 FS2 F3 and in general there would
be capital J number of such objective
functions I'm not mentioning anything
for this example about the decision
variables they can be l in number L can
be and that's capital l l can be can be
three can be four can be five can be 10
accordingly now the bounded area which
you'll see now would basically depict
the hypothetical fible set and the
boundary would be considered to be
Optimum set of feasible points depending
on whether it is a maximization or a
minimization of that objective function
which you are
considering now here it is important if
you remember for the other examples when
we considering those X1 star X2 star X3
star using the different colors and the
values of the objective function Capital
fub1 F2 F3 I was always mentioning the
point that you are trying to maximize
hence two being less than three would
mean that the objective function value
is better to have three than two so we
are always trying to maximize but it can
be the other way around also that means
we want to minimize miniz so the
minimization concept and the ideas what
I mentioned in the last slides would
continue to hold true but you have to
look at the point that you are trying to
basically reduce that value of the
functional form that is fub1 F2
F3 now in the coming uh set of of the
diagrams there are different type of
color combinations which I used in order
to make it clear the red colored would
basically the parro optimal points in a
b is for the case when we consider the
the minimization of objective Function
One along with the minimization
objective function two so if you notice
the second bullet point I did mention
there is a minimization maximization
Conant also so in combination there
would be four different ways of trying
to analiz which is maximization of fub1
which is objective Function One along
with maximization objective function two
then minimization of fub1 and
minimization FS2 then maximization of F1
minimization of F2 and the last one will
be minimization on F1 and maximization
on F2 so with this I will close uh this
uh 23rd lecture and continue this slide
would also be shown in the next class or
in the next lecture because this will
continue with some discussions and have
a nice day and thank you very much for
your
[Music]
attention
[Music]
[Music]
[Music]
