Lecture 45

IIT Kanpur July 201
20 followers
Mar 17, 2024
[Music]
welcome back my dear friends
participants and students for this
multicriteria decision making course
which is under NL Muk series and as you
know this is spread over 12 weeks with
60 lectures and each each week we have
five lectures each lecture being for
half an hour and my good name is
raghunandan s Gupta from the IM
department at iport so if you remember
for the last uh few classes about two
weeks we are discussing about the multi
criteria decision making where the
decisions Alternatives criterias are
more subjective not objective as like in
m modm and we were discussing
concepts of topsis method and we are in
the process of discussing the electri
method and in the electri method
obviously you'll see in the slides we
are discussing about the different type
of distance measures like the Manhattan
distance the L Infinity then L2 which is
the general distance measure between two
points the coordinate system we did
mention about maharan distance Hamming
distance but all our Focus would be in
different uh LP Norm or the lp distances
considering the Manhattan at one end and
the end Infinity distance at the other
end and this is the 45th lecture which
is the last lecture in the n9th week
broader umbrella as I keep repeating is
basically mcdm we have already covered M
modm few of the techniques we are into
multiattribute decision making
and the different techniques like as I
mentioned few minutes back topsis
Electra HP Vore and all these things so
the coverage is where we started and
left in the last class was Electra
method which is elimination and choice
translation
reality now if you remember we had drawn
and
considered uh the two different points X
and Y if you follow the last lecture
you'll understand X and Y and they are
then three dimension like X is a vector
which has X1 X2 X3 Y is a vector which
is has y1 Y2 Y3 other the coordinates
and based on that we found out what was
the distance based on Manhattan Norm
then L2 L3 till L infinity and we
plotted that if you remember that and I
showed you that for manat Norm it was
the Max and infinity it was the
minimum now further
um analysis of that I have drawn this
graph where the same ideas has been
converted into the concept of different
coloring and if you notice there is a
red color blue color green color and
yellow color now if I consider first the
green one so let us consider two
different
points I'll mark
them as a here B here and if if I have
the coordinates this is a two
dimensional one so it's easy for us to
visualize so you have the coordinate
system for a r X1
y1 for b r X2 Y2 obviously I can find
out the distance between A and B which
is based on square
root of X2 - X1
s + Y 2 - y y 1 whole
square and we take the square root so
that would be given by the green line
which I'll put up Mark in the this one
I'll
just again redraw it so this is the I'll
hash it so there's no
confusion this is the L2 now if you see
the red blue and yellow they are as
follows the green line which I just
hashed is the unique shortest path while
the red blue and yellow are based on the
L1 norm and are of the same length but
you can take different parts why of the
same length so if you consider the total
distance covered same length means the
L1 distance of the same length so if I
consider the yellow one so it goes I'm
not comparing with the L2 I'm only
comparing the
L1 so if I take the yellow one so I
go 1 2 3 4 5 units right then go one
vertically up which is 5 + 1 then I go
right +
1 and then then again go up 1 2 so it
will be going 1 2 3 4
if I see the red one so red one
basically I would go 1 2 3 4 5 6 and
onto the right I go 1 2 3 4 5
6 so let me count if I taken it
correctly 1 2 3 4 5 6 1 2 3 4 5 6 so 6 +
6 and in the yellow one let me count if
I missed
something um 1 2 3 4
5 plus oh it goes two up sorry my
mistake it goes two up then goes right
one then goes up vertical again 1 2 3 4
so the total amount of movement if I
follow the L1 basic concept if I Follow
the yellow path it is 5 + 2 7 7 + 1 8 8
+ 4 is
12 so yellow is difficult for you to see
but it's 12 number of
units and if I consider the red one also
it is 12 units movement so the L1 n
different it can have different parts
but the distance covered is
same so that's why the L2 Norm has
unique solution while L1 Norm does not
have any unique solution lengths are
same but you can take different paths
one can generalize this in the nend
dimensional one we did it in a two
dimension we can have in a three
dimension like Corner points or some
points in a cube and I can find out the
distance based on L2 and the
L1 now cons why did we discuss the
distance it will be clear later on now
the ideas if you remember that we were
discussing the steps of the Electra
method and in the the concepts of what
we did in
the topsis method so this these Excel
sheets which is uh named as mcdm and
there are different sheets inside the
Excel sheet the first one I have already
sent the for the eth and the nth class
I'll send along with the slides the
Excel sheets would also be shared with
the participants and the
students now why did I go back to the
Excel sheet if you remember we were disc
discussing about the distance based on
pis and
NIS and I categorically mentioned there
the pis and the ni distance were based
on
L2 so the idea was that was can we
convert that into any other distance
concept yes it can we will see that in
the Electra method so Electra method
sheet is still empty I'll fill it up
with the discussions as we proceed now
same step set of steps so the steps
which we had followed in the topsis
method was first have the X Matrix X
Matrix is based on the number of
Alternatives and altern of criterias the
Alternatives were marked along the rows
criteria along the
columns then they were normalized
normalization is based on the different
type of utility functions or
normalization concept we used and I did
show in the Excel sheet and and it is
also there in the video which we which I
was
delivering that normalization was based
on simple normalization like x i j if I
consider each value in The Matrix x
capital x which is Matrix as xig XI J
divided by the sum of all the xes
depending on whether you are trying to
normalize along the rows or the columns
or you can take the logarithmic value or
you can take the square values so in the
Square values it will be square of xig J
divided by the sum of the squares along
the rows of the columns in the
logarithmic scale when you're
normalizing it will be log of xig
divided by sum of logs of xig along the
rows and the columns we'll follow the
same concept here
also so here we convert the entries in
the decision Matrix and normalize them
now in in the slides and in the Excel
sheet or in the discussion generally the
concept of trying to convert X X into
the normaliz vector multiply the weights
weights are based on the criteria only
then proceed accordingly the concepts
and the methodology remain the same may
be the nomenclature of what you want to
mention as X or a normalized vector that
symbol would be different but there
should not be any confusion so you
convert the entries in the decision
Matrix X or a a into a scale normalized
value so this would be square
so I have aigs xig would be AIG squ
divided by the sum so this should not be
a square root it should be just be the
sum of a kg
squ and when I
mention about the squares So based on
that I find out the the
normalized values along the rows of the
columns now what are
these These are on a scale SC factor of
one or 100 If I multiply the scale
factor by 100 it becomes on a 100 scale
I'm trying to find out the level of
importance because the utility function
based on which the person is analyzing
the points of the values for each and
every criteria for each and every um
Alternatives need to be scaled and found
at the same level such that comparison
becomes easier so once we get the
normalized vector we should remember
again the point which I'm mentioning
repeating please bear with me here M
would be the number of Alternatives n
will be the number of criteria so here
if I read the second last bullet point
here m is the number of Alternatives
shown along the rows and N is the number
of criteria shown along the
columns the normalization which you do
implicitly depends on the utility
functions or the worth which the
decision decision maker thinks what is
the value which accurs to each and every
Alternatives based on the criteria what
he or she is
analizing so say for example the car
example so the car are there are M
number of Alternatives in front of you
the criteria can be I'm just repeating
can be the mileage can be the cost can
be the maintenance cost can be the
safety features can be the color can be
the luggage space can be the number of
passengers who can be carried it can be
the resale value So based on that the
points are given if I consider the
example of college choosing a college it
can be what is the tuition fees how far
it is from home it can be what is the
quality of the education what are the
job facilities internship what are the
final job facilities how are the
teachers what is the environment and all
these things can be considered similarly
we can consider different ideas when
somebody's buying an apartment I keep
my discussion limited to these three
because they can be easily communicated
with the participants and the
students now here let us consider so the
normalization has to change because it
should be not the square root a square
here so let me do the problem
accordingly so I'll consider of M as 3 n
as
3 so let me Mark
it
as A1 A2
A3 this is
C1
C2 I should uh Zoom it more so it's
visible to
all and
C3 and this is Example The
Matrix uh
X and I would also have the Matrix of
Weights so they would be corresponding
I'll fill it up just
C3 and there were also
C1 C2
C3 and the values are given based on 2 1
2 then I have 3 3 4 1 I'm taking the
same values 1 3
2 now I
write the normalized vector as see for
example
R so here also I have A1 A2
A3 so the corresponding criterias
come so now I normalize so let me use
first so if you if I go back the
normalizations are based on the
columns so if I consider first
example I'm normalizing base based on
just xig divided by the sum so it will
be sum of along
the along the um
columns so I need to fix the sum value
because sum value along the column
doesn't
change so I find it out and it is the
sum I want to
verify so it is
is equal to one so I highlight it with
the color so it's easy for us to
follow so now
I
take
as
D to D so D column again
normalized sum is equal to
1 I go to the third one mean
C3 so it will be sum of
e so I sum them
up and the value comes out to one
perfect so this
is I
should Mark this
with some orange color here so there
it's becomes visually much
more
noticeable and this sum is
yellow and
uh this was the
weights so I
should shift it one box so it's easier
so this is the
weights I also Mark it with light orange
so the so we need to find out the
weights So based on this I find out
the weights so same concept once the
weights are found out I need to multiply
the normalized vector along the weights
weights remember are based on the
criterias only what is the level of
importance of the criteria which are in
some way the constraints because
alternatives are the objectives which
you want to choose car you want to
choose buy the house check with the
college which college you want to get
admission and once the weights are given
based on the decision maker we will
multiply The normalized Matrix with the
weights Matrix only thing to be
remembered which I did mention twice be
aware that what is the dimension of w
what is the dimension of X So based on
that you multiply so step two would be
waiting the normalized decision
Matrix which will do so consider the
weights are 1/4 2/4 1/4 so is
1/4 24 24 is 50% and 1/4 so double check
the values of25 +5 +5 should be one so
the other values are all
zero so I have done this so let me add
few rows so the
H so what I did was I
will
insert so I found out from X to R so
I'll put Arrow here this is the
calculation which we
doing so I'll put the
arrows in black color so this is the
calculation then I need to multiply the
weights and the normalized vector R so
I'll put the arrows accordingly so but
first let me put
the Matrix it will be mm U
LT
of the first
array into the second
array I have all the values so these are
the corresponding values so once I do
that so these I'll basically mention it
as y
and then Mark them
with red values and bold them so based
on that we'll do the calculations so
what we do
is multiply the
weight with the normalized
vector this we
do and we get y Vector so I put the
arrows accordingly in the same way if
you remember in the tops is also I put
the
arrows so that
was to
avoid confusion so once you
have y so I would not put any color
here so this is
y and these are the cell values for the
W which I'll put as blue in the same way
as I did for the topsis
so remember the weights is one if you
think or perceive that for criteria two
you'll assign two times more weightages
than criteria one now this is a point in
in qualitative context I'm mentioning it
will become clear when we consider more
of these other different techniques so
twice more importance Thrice more
importance 10% less import based on the
criteria so here it is mentioned
criteria is twice as important
while for criteria 3 the level of
importance are the same as criteria 1 so
if criteria 1 criteria 3 are same
obviously criteria two would be twice
with respect to one twice with respect
to three So based on that the weights
are there I immediately put on the weits
but the qualitative statement would be
like this what is the percentage what is
the importance what is the level of
importance how many
times so once the weights are given I
multiplied and find out y so the values
are given so the values are what we have
calculated in the cell so this set of
point I'm only reading till three places
of decimals
8362 1
0125
25050
041018 7
01
0 0.10 0 so these are Y which in this
even though different because the Y
calculations are different they are
based on this
now we'll take a pause if you remember
in the topsis method we considered the
pis and the N the distance from the most
positive and most negative here the idea
was if you remember the concordant set
and discordance set liking set disliking
set now here also I mention one point in
details that liking set and disliking
set doesn't mean that uh we are only
taking one alternative so considering
there are two Alternatives A1 and A2 for
A1 I find out the liking disliking set
for A2 also if I find out the liking set
disliking set so what does it mean to
repeat consider I have taken alternative
A1 so obviously there are some positive
points but obviously it can be possible
that they can be NE negative points on
two accounts negative points based on
the account that there are some
disbenefits acre for alternative A1
consider buy the
car very good car but the maintenance
cost is very high so that is a
disbenefit or consider the college
College is very good but cost of
studying is very high so that is the
disbenefit similarly if I take the other
alternative they can be disbenefit also
benefit see for example for buying a
very uh less costly car that is a
benefit I'm paying
less but uh the disbenefit may be
longevity of the life or later on the
maintenance cost exponentially increases
because for wear and tier or the road
road conditions are not good the car has
to maintain much more often so we will
consider this levels of benefit and
disbenefit so we need to determine the
concordance and discordant indices sets
based on the in IND set we'll formulate
the concordance Matrix and the
discordance Matrix now what is the
concept of concordance and discordance
what I mentioned I'll repeat it
concordance index is defined as the
amount of evidence to support the
conclusion that alternative AI out ranks
dominates or it's better than a A
J now that is the index so obviously I
have to compare that based on each and
every criteria similarly what does the
discording index means is the
counterpart of that and it in very
simple mathematical terms one can say
that is the discordant index is the
complement of the concordance index
based on the dis benefit which I'm
getting by choosing AI over
a now the distance concept ccept and
here the distance concept I'll slightly
go into a different analysis of trying
to analyze the
distance now when we are discussing the
liking disliking consider this
example and I'll give you a mathematical
background of that also so if I consider
first the mathematical background I'll
draw it here it has nothing to do with
the problem solving is basically to do
with the
analysis so consider I have access X and
Y and I'm trying to find out or estimate
in simple statistics so if I estimate I
need to find out the variance if you
remember the variance is square
quadratic in form as per the formula so
if I put the concept of quadratic
equation for trying out to find out the
variance the variance function looks
quad
like this now why I do it if I
plot the
value of delta what is Delta Delta is
the difference between the actual value
and the estimated value parameter and is
estimate so this point which is the
origin where both the estimator and the
parameter exactly match so Delta is zero
onto the
right it on as I draw it is positive on
to the left is a negative so if I
consider a parameter value of say for
example four so in one case we estimate
four by a value of plus two in a another
case we estimate for by a value of say
for example minus 2 so in this case oh
consider for for I'll give you an
example +
two -
2 let me take the value of
wait I'll take a much better example
wait consider the value as Theta which I
want to find out and Theta hat I'm using
a symbol as the
estimate so difference
between Theta hat minus Theta is Delta
which I'm
measuring I will come to the values of
the parameter and estimate later on
first let me come to the concept of
Delta only consider in one case Theta
hat minus Theta is + 2 and another case
the difference between Theta hatus Theta
is
min-2 so + 2 whole s is 4 - 2 whole
square is four so if I basically try to
find out the the quadratic value so in
both the cases for negative and positive
was of same equal distance onto the
right and the
left but the answer is that when I come
to the analysis here is it true answer
is no why not let me give you two
examples before I
proceed consider an example where and in
concept of a real consider an analysis I
win or I find out 100 rupees and in
another case I lose 100 rupees this is a
very simple and rudimentary example but
try to understand that so my net worth
of the value of the satisfaction of
winning 100 rupees with respect to the
value of minus 100 rupees being a loss
my level of dissatisfaction that I've
lost 100 rupees would be much more than
winning 100 rupees so my loss I'm
quantifying much at a much higher level
even though 100 rupees losing and
gaining worthwhile is the same in the
units consider another example and here
I am trying to basically say for example
if I consider from a civil engineering
perspective I want to build a dam the
Dam's height is basically actually
should be 120 ft but consider there are
two instances in one case the Dam's
height is 122 and it has been built and
another case it has been built to 118 so
what is the difference between 120 and
122 is in one case if I consider it's
plus two another another case 120 118 is
min-2 if I Square them up considering
the quadratic of the concept of variance
both case plus 2 whole square is 4 - 2
whole square is 4 so is equally
penalized but the question is that in
the practical sense is it right the
answer is no wrong it's not right why in
the case if I build it to 122 ft the
there would be some extra cost Manpower
extra materials extra time but why want
122 has been decided because on an
average the flood comes and the level of
the water Max can reach 120 so in the
case if it is built on 122 then the
probability of the water breaching the
dam is much much less now the answer is
much much less with respect to what
consider if it is built to 118 and the
flood comes and the average height of
the water is 120 in that case the
probability of it breaching the dam is
much higher so initially when you build
it to 118 the amount of Manpower amount
of material amount of time taken was
much much less but the catastrophy loss
it will be happen due to the environment
impact of the flood overrunning and and
and breaching the dam would be much more
so in that case here for this example
overestimation of 120 by 122 with
respect to 120 and 118 overestimation
would not be penalized underestimation
will be penalized and there can be other
examples of of overestimation being
penalized underestimation not being
penalized and so on and so forth so with
this example and with this thing in the
mind we'll discuss the problems
accordingly thank you very much and have
a nice
[Music]
day
[Music]
[Music]
