noc19-mg15 Lec 25-AHP

IIT Kanpur July 201
20 followers
Dec 3, 2023
AHP, Normalization, Consistency Ratio, ELECTRE process, Liking set and Disliking set, Outranking
Welcome back my dear friends and dear students, a very good morning, good afternoon, good
evening to all of you wherever you are and as you know this is the DADM 2 which is decision
data analysis and decision making 2 course under the NPTEL MOOC series.
And this total course duration is for 30 hours which is for 60 lectures spread over 12 weeks
and each week we have 5 lectures each lecture being for a half an hour and we are in the
5th week and as you can see from the slide which is the 25th lecture which is in the
last lecture for the 5th week.
And my name is Raghunandan Sengupta from NIME department at IIT Kanpur in India.
So we were discussing the AHP ranking of buying a car and when you when you are taking a decision
if you remember that we had considered first the subjective criteria the style, fuel economy,
cost and so on and so forth and then you first compare amongst themselves make a priority
vector based on the priority matrix.
Then you take each and every criteria for all the alternatives combined together one
at a time then also you have separate priority vectors combine this priority vectors for
both the levels and both the hierarchy multiply to get the score.
Now if the cost factor comes as the fuel economy was there kilometers per liter that can also
be normalized using the same concept on normalization which I have which I am repeating it time
and again that can also be considered and that can be added on to find out the final
ranking.
Now if the cost structure comes I said that adding or subtracting one of the criterias
or one of the not all the alternatives but the criterias may change the relative ranking
of the criterias like I bringing a fourth one in this case or say for example adding
say for example deleting of the any arbitrarily third one provided there are more than two
the ranking may change relative ranking.
So here also you consider the cost structure and they can be different flavors of the problem
remember that and you can basically find out the relative scores.
So continuing with this once you have the combined scores so Scott is the winner with
the highest benefit to cost ratio hence it is in the first position combine all the scores
second position will go to I20 third to alto while the fourth position goes to civic.
So obviously at this position depending on the alt the criterias which I have already
mentioned and the cost structure you have this.
So the problem concept is very simple you have to find out the priority vector priority
matrix priority vectors rank them combined one level at a time and also told that they
can be primary level tertiary level then below the tertiary level one level two levels below
the tertiary level and he can go on such that you add up the scores as you go up and keep
multiplying and adding them such that you give a final scores to rank them.
Now obviously there are some pros and cons for AHP so it allows multi-criteria decision
making which is a very good point.
So multi-criteria with different types of one subjective objective concepts can be rationalized
and obviously you have seen that consistency ratio consistency values the other scores
will give you how consistency concept can be utilized in trying to analyze how rational
your decisions are so that can be utilized.
So AHP is applicable where it is difficult to formulate criteria evaluation that is trying
to basically combine qualitative and quantitative is much more rational using AHP if rationality
definitely would be considered we can add them accordingly.
It is applicable for good group decisions so when there are more than one decision maker
and the decision has different type of alternatives different type of criterias like if you remember
a very very nice example though not solved in details because it was very simple problem
Ram and Shyam making a decision to go to the MBA schools and then their father come.
So at each level there are many decision makers and then you combine say for example you are
trying to analyze the problem where you have say for example I have mentioned it very briefly.
So say for example candidates has applied for the HR general manager or HR senior manager
and on the other side of the table the MD, the chairman, the vice presidents like say
for example 5, 6 persons are sitting and they want to analyze how the good or bad the
candidate is.
So you will try to basically analyze the candidates academic credential what type of work experience
he has and what is his psychological profile and what is his demand of salary and all these
things would be considered but all the people who are trying to analyze the candidate will
have different criterias different levels of scores for the same set of criterias.
So combining them would give you a much more better perspective.
Now when you combining for the case of Ram and Shyam the parents were giving equal weightages
to their sons Ram and Shyam it may be possible the chairman and the MD has a higher say hence
technically the scores which you will assign for the overall ranking coming from the point
of view of the chairman the MD would be higher while the general manager and the vice presidents
who are trying to analyze the candidate their score would be low.
So which would be a much more rational decision process when you combine all the alternatives
all the criterias and take subjectivity into consideration.
What are the disadvantages?
So there are hidden assumptions about consistency and consistency trying to find out level consistence
to be equal for all the decisions is difficult because it becomes very difficult to for me
maybe it is very difficult when I am trying to analyze the decisions on a one-to-one basis
that means i is to j and forgetting about all other criterias or all other alternatives
which are there.
So I only consider an i and j but when I come to j and k I only consider on j and k so trying
to combine and then when I come back to i and k then j is removed.
So trying to combine i, j, j, k and i, k, i, j, k are say for example the decisions
or the alternatives whatever it is or the criterias and then trying to basically combine
them and find a related ranking between i, j, k may not give us a very good consistency
ratio.
So this has to be thought about.
Repeating evaluations is cumbersome because repeating the same thing time and again may
not give us the same relative scores which we actually want to have.
So if you remember the principle diagonals one there is no problem but trying to basically
combine for civic and I-20 on different fronts that the differences of the scores may be
very vague if you keep repeating that.
So hence the actual scoring system which should be between the cars or between the criteria
may not come out when any person is analyzing or it may be possible they say for example
when buying a car the first person analyzes and gives the score that the second person
when he or she wants to gives his or her score he or she may be influenced by the decision
of person 1.
So those influences we consider are not there but it may be possible those decisions are
there.
Say for example in the other example when trying to come back find out how would you
recruit senior manager for the HR.
So the chairman may have a higher say and his or her saying may have an influence on
the vice president on the general manager he is also trying to analyze and they may
be influenced.
It is difficult to use when the number of criteria alternatives is high it is more than
7 because trying to compare 7 see 2 combinations because taking 2 combinations from 7 set becomes
very difficult and very confusing when you keep repeating this comparison.
It is definitely difficult to add a new criteria in an alternative because the relative ranking
of the alternatives and criteria would change absolutely true.
This is one of the most difficult points for AHP which does not give it a positive light
obviously it is on the negative side but still AHP is used quite heavily.
And it is also difficult to take out an existing criteria alternative since the best already
might differ when you compare them against the existing alternatives when one of them
has already been taken out.
So the relative ranking would suddenly change whether you add that or you take it out as
I mentioned if you have say for example 7 different alternatives I am considering in
a very arbitrary sense and if you take out the one of them then the remaining 6 ranking
may change relative ranking or if you basically have 7 one and you add the 8th one then I
also the relative ranking may change because the scoring pattern which you are doing for
even the same person would obviously change quite drastically.
Now I will just go through some brief very simple mathematics this is just for information.
So this I thought I should have covered in the initial part but when I was when I was
making the slides I thought that best for AHP is to discuss the concept go directly
in the problem and then come back to the simple concept of consistency which I had been repeating
time and again today also that means in the 25th lecture and also in the 24th lecture.
Consistency implies coherency in the judgment process.
So there is coherent concepts or coherency of logic logical sequences of steps are there
when you make a decision whoever it is making a decision buying a car choosing a house or
trying to basically recruit any person in your company and basically the coherency should
be there in the judgment process as I just mentioned for the decision maker but given
the problem in human judgment the property of consistency may not always hold true so
that may change.
So that is why we do the consistency check through the ratios few of the important properties
for consistencies are.
Now for consistency we would always have the if you remember the i, j, k is this suffix
which were there and the numbering was depending on the number of criteria number alternatives.
So if you have i, j, k running from 1 to n and if you find out the multiplication so
if you remember I multiplied the matrix with the vectors to give the overall ranking.
So if you multiply this a, i, j cell with the a, j, k cell the values gives us the total
number of such scores addition of this of the score.
So you what you are doing is you are adding up the values i, j, k, a, j, k for all values
of j. So that will give you one value which will be in the vector of the final score if
you remember we have done that. So those doing that number of times obviously will give you
a different levels of consistency. For a 2 by 2 matrix it is always consistent we would
not go into the proof but I will state it.
So consistency implies that all rows and columns are linearly independent. So it is something
to do with matrix a linear set of equations. If you remember the Gauss Jordan method of
elimination we are considering that there are n number of equations and n number of
variables and each of them are independent all equations independent of each other hence
you will get unique solutions. So if you basically have the matrix concept so here all the rows
or all the columns are independent and none of them can be expressed as a linear combination
of the other.
Hence the concept of rank will come out of the matrix as that the rank would be n depending
on the number of rows or number of columns which you have. Now the matrix is consistent
so consider it is of the form that you have the weights given as w1, w2, w3. So these
are the weights for the first one is w1 second one is w2 similarly for the last one is wn
and the values which you have for the priority matrix obviously as usual the principal diagonal
is 1 and the of the diagonal element which you have these are not symmetric.
So a21 is not equal to a12 because if a12 is 9 a21 will be 1 by 9 or if a21 is 7 a12
will be 1 by 7. Similarly if I consider an to a2n again they are not symmetric because
the scoring pattern which you are doing for the criteria and the alternatives are such
that you follow the same scheme of scores of 1, 3, 5, 7, 9 and if this ambiguity you
bring the even numbers. So in this case it is an asymmetric matrix the principal diagonal
has been 1 since the priority vector. So if I find out the weights so weights would basically
be the ratio of the sum of all the corresponding rows of the columns and if I have them the
correspondingly the weights w1 or w2 or w3 will be found out accordingly and we can find
it.
So this is just our end result I am not going to go into the actual mathematical derivation
and the use of that but remember one thing consistency property is important for HP 0.1
the ranking system would change if you add or delete particular criteria or alternative
that is also important trying to find out the priority matrix you have to have the level
of consistency judgment falling up in the rational front such that consistency ratios
or consistency concepts are not violated in the final decision given the consistency matrix
you will try to find out the consistency vector so sorry the priority matrix you will find
out the priority vectors and go from the lowest bottom level from the tertiary to the primary
and so on and so forth to give the cumulative scores where you find out the sum of the multiplicity
factors of the scores.
And another thing which I have repeated time and again so remember that the utility function
which you follow or normalizing whether through the rows or the columns such that the sum
is equal to 1 after normalization the normalization concept of whether it is quadratic or exponential
or power function or logarithmic function whatever you are utilizing would be used consistently
for the same person for all the alternatives and decisions and try to use the same normalization
concept of the utility function for all the different persons when you are trying to combine
the decision makers like decision maker 1 or 2 or 3 or 4 when trying to arrive as a
collective decision like say for example for the car or say for example trying to recruit
a person the example which had given.
So with this I will end this AHP part and then start of a new concept of the Electra
process so this technically I thought as we are going in the flow it would have ended
in the 25th lecture and end of that but I have a finish a little bit ahead of time.
So whatever time is left for this 25th lecture which is the last lecture for the 5th week
I will consider a new important topic to be added. So just bear with me I will open the
slides and discuss that accordingly.
So this is the AHP part and just one thing so finally to wrap it up for AHP so then once
we find out sorry for that I just missed the slide so when we find out the actual A matrix
so they would be normalized so and finding the ratio so once you find on the ratios hence
the matrix A which is the normalized concept you multiply the weights is it will be equal
to n into W and if and only if A is consistent and in this case if they are inconsistent
you can find it out but for the consistent scores you will use the average.
If you remember in the first example when we are doing the consistency ratio we found
out the averages and proceeded accordingly. So these are just for interest you can pick
up the Satish book the references which are already given when I started the class so
you can check that and find out the actual mathematical formulation of consistency how
it can be utilized. Here in this course we remember we are just discussing the application
of different type of MCDM processes.
So with this I will end I will again repeating end AHP and then I will try to start so we
are continuing the 25th lecture so just for when if at all I upload the slides so I thought
they should not be in confusion okay. So we are going to start a new method which is known
as Electra which is elimination and choice translation reality. Now I will try to finish
all this stand alone topics in such a way so this continuity without this break which
happened in this fag end of the 25th lecture obviously there is no confusion but to make
a smooth flow it will be much more easier for you all to understand and easy for me
to continue without any break in the flow.
So what does Electra mean? So Electra is basically the word is a French word which basically
means elimination and choice translation reality. So you are doing the concept in such
a way that you have a set of criterias and decision and you try to find out the best
possible decision which is a theoretical concept such that you will try to find out the positive
distance of liking from that the so called theoretical set or theoretical value for all
the decision making this is an alternatives which are there and also try to find out the
negative weights that means the negative distance from that so called theoretical value for
all the decisions which you are forced to take due to some circumstances.
If you remember I did mention one of the example that you are a bank and you are forced to
or due to government regulations you are told to open it in the rural district. So if you
consider that profit motive is your important factor obviously that is not the best choice
but still you have to do that so how do you make a decisions accordingly. So you will
consider some distant matrix coming out such that you will divide the overall decisions
into two sets number one is a positive set which gives us positive values and a number
two is the negative set which gives us negative values and then try to compare and combine
this positive set and the negative set and try to basically take a collective decision.
Now the point of positive set and negative set is in this way set means a set of scores
which you are giving. Now the point of positive and negative sets in this way say for example
there are two alternatives i and j. Now we will consider as we have done in the HP that
you give a score of 9 is to 1 9, 5 is to 1 5. So the point of 9 or 5 or 7 or 4 whatever
it is you are giving to the case of the decision alternative which is positive to you and you
give a inverse of the score to a one which is definitely not positive to you but you
are being forced to take the decision due to some circumstances.
In the seminal way we exactly do this we basically compare two decisions two alternatives i and
j and if i is better we give a positive score of positive point to i and when if you are
forced to take j we also give a score but the point is that you try to rationalize the
score in the normalized scale that i is to j and j is to i if the points are equal which
means that I am equally disposed whether I take decision i or decision j and if I am
not equally disposed between i and j say for example I like i more than j or I like j more
than i in that case the preponderance of the scores between i and j and j and i if I like
i and if I like j more than i then the preponderance of the score of j to i and i to j would be
such that the points of j would be higher.
This would be coming up correspondingly using some very simple concept of liking set and
disliking set. Here remember in the initial case we will consider the liking and the disliking
set to be such to be made in such a way that the distant matrix are of equal proportions
in the sense if I like something my propensity of distance would be of same magnitude at
as to the level when I dislike something which may not be true because in the actual sense
if you remember I have discussed the concept of asymmetric loss functions when I am assigning
a higher score or lower score or a higher loss or lowest loss depending on whether my
estimation is high or low if you remember I have considered three very simple examples.
One was basically for building the dam from the civil engineering concept one was basically
from the electrical engineering where you have a set of a machine is there and you have
vacuum circuit breakers to understand that when you should basically do the general maintenance
of that very costly machine.
And another example was that when you are trying to float in a product in the market
from the market perspective the warranty life it is high or low would basically help
you to gain the market in initial case but you lose the market as your products fails
earlier or in the second case that you are not able to win the market in the initial
case but you will slowly gain the market as your products being popular considering the
warranty time is actually different with respect to the competitors.
So these three examples gives gave us the concept that whether over estimation important
for the case when building the dam under estimation important for the case of the
electrical circuit and where case over estimation or under estimation can whatever is important
for the marketing example.
So this concept initially would we would not use in the electro and but will be try to
bring that using the distance concept.
In course of time six different okay let I have not finished the first bullet point.
So this is the elimination and choice translation reality which is in English it was developed
by a Frenchman by Roy in 1990s when trying to basically analyze different huge projects
where decisions were both subjective and objective to arrive at a rational decision in trying
to rank projects where different criteria's different alternatives were utilized.
It is one of the MCDM tools based on the concept of outranking method that means how good or
bad your ranking system is how you are able to outrank other decisions or other alternatives.
In course of time six different variants of Electra came into the picture.
They are Electra 1, 2, 3, 4, Tri and IS methodology and they have been developed where each by
has some unique properties and hence variations when compared to each other would be there
but you will only consider the simple concept of Electra how it is utilized.
A good text related to comparison of these six different Electra method can be found
by the in who is was basically the inventor Roy in 1990 and one paper 1996 and by Rogers
the paper was out in 2000.
For our understanding we concentrate on the basic concepts of Electra 1 process only in
a very simplistic sense but as I said I will add the concept of outranking in asymmetric
sense also like whether I like or dislike the weights would be different depending on
the level of loss function or the asymmetricity which you want to bring into the picture.
And afterwards we will give comparison of each of the six different variants for the
appreciation of the readers or appreciation of the students who are doing this course
but I will mean concentrate on the Electra method with outranking concept being equally
disposed whether you like or not like and then being unequally disposed using the asymmetric
concept of loss.
The concept wise we will bring it very simply.
The first question which invariably comes to into one mind is what do we understand
when the concept of outranking.
So, what does outranking mean does it is does it mean a relative score does it give us the
absolute score what does it mean.
Stated very simply the method outranking is the idea in which one compares one decision
or alternatives against all others and ranks them based on some set principle and as you
do that you are trying to basically find out the relative score where the decision alternate
stands with respect to the all the other sets of decisions and alternatives.
In doing so one assures that at the end of the pair wise ranking here also it is a pair
wise ranking.
So, if there are n such alternative decisions you take the jth one and compare the j to
1, j to 2, j to 3 so on and so forth and then you go to compare j to j minus 1, j to j plus
1 and go till j to n and put j alternative decision in the right perspective where it
stands.
So, let me continue reading it in doing so one assures that at the end of the pair wise
ranking process we have a unique ranking system amongst all the decisions on alternatives
such that we are able to take the decision correspondingly.
Now nomenclature would be important so I will go a little bit slow.
Thus intuitively consider you have the alternatives A given as A1 to An so each alternative will
have criterias so it can be C1 to Cm for A1 then again C1 to Cm for A2 and so on and so
forth.
Now the number of m which will have for A1, A2, A3 need not be same so if they are not
the same so obviously we will consider some of the cells to be blank or replace them accordingly
in order to have a consistency concept of ranking.
This word of consistency I am using not from the AHP point of view just from the English
language point of view.
So where the number of alternatives is given as n such that for each alternative they would
be not m but we are considering l number of criterias.
So when we are comparing AI1 to AI2, AI1 means the suffix is I1 is basically the I1
is one element of n and I2 is also an element of n.
When I am considering alternatives 1 to 2 or 1 to 3 or 1 to 4 like I said I1 to I2 then
I will consider each and every criteria to be considered such that I am able to compare
using criteria 1 I am able to compare AI1 and AI2 then considering criteria 2 and again
I am able to compare AI1 and AI2 and continue doing that such that I assign scores at each
level of the distance and then do the collective combination of the scores.
So it says that these 1 to l are the set of criterias and one accomplishes his or her ranking
based on the collective the cumulative effect of all these l criterias which I just mentioned
when comparing two different alternatives one at a time and or say for example considering
the other things accordingly.
At the end of the comparison process we end up with the best choice set where we say that
A1 or A2 or A3 are ranked in such a way based on the criterias we can rank them accordingly.
So with this I will end the last lecture of the fifth week and continue the discussion
of electro in more details and solve a problem in the same concept solve a problem and then
come to the actual mathematical algorithm of decision process how you make that.
So have a nice day and thank you very much for your attention.
