multi dimensional ranking
TRANSCRIPT
![Page 1: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/1.jpg)
ON METHODS FORMULTIDIMENSIONAL RANKING
By Shrinivas Vasala
![Page 2: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/2.jpg)
OBJECTIVE:
To Study different ranking methods
Implementing these method on agiven data set.
![Page 3: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/3.jpg)
THE MULTIDIMENSIONALRANKING PROBLEM
m candidates (or “alternatives”) M = {1,…,m}: set of candidates n criteria (or “agents” or “judges”)N = {1,…,n}: set of criteria Each voter i, has an ranking i on Mi(a) < i(b) means i-th voter prefers a to b
The rank aggregation problem:Combine 1,…,n into a single ranking on M, whichrepresents the “social choice” of the voters. Rankaggregation function: f(1,…,n) =
![Page 4: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/4.jpg)
METHODS
For finding thewe rank
bestuse with
optionsome
fromsound
all the feasiblealternatives,aggregate attributes.
approaches toonerespect to more than
Positional Rank Aggregation Methods
TOPSIS METHOD (Technique for Orderby Similarity to an Ideal Solution)
Preference
KEMENEY’S METHOD
![Page 5: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/5.jpg)
POSITIONAL RANK AGGREGATIONMETHODS
Pluralityscore(a) = # of voters who chose a as #1: order candidates by decreasing scores
Top-k approvalscore(a) = # of voters who chose a as one: order candidates by decreasing scores
Borda’s rule [Borda, 1770]score(a) = i i(a)
=[sum of rank value]: order candidates by increasing scores
of the top k
![Page 6: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/6.jpg)
TOPSIS METHOD[HWANG AND YOON, 1981]
The factors involved are: Ideal and Negative ideal location Distances from Ideal and negative Weights of item Sum of square of each items.
ideal location.
Relative closeness of Ideal and Negative ideallocation.
![Page 7: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/7.jpg)
EXAMPLE:”A SURVEY FOR DEMAND DRAFTCHARGES IN INDIA”
S.No Bank name Category avg <1K avg 1-5K avg 5-10Kavg >10K (for
every additional 1K)
Cancell- ation fees
1 Allahabad Bank Public 25 35 40 3 1002 Andhra Bank Public 22.5 25 27.5 2 1003 Bank of Baroda Public 20 35 35 3 754 Bank of India Public 32.5 42.5 42.5 4 52.55 Bank of Maharashtra Public 25 30 35 3 1006 Canara Bank Public 25 40 40 3.25 5257 Central Bank of India Public 22.5 32.5 32.5 3.5 758 Corp. Bank Public 30 30 40 2.5 1009 Dena Bank Public 35 35 35 3.25 85
10 Indian Bank Public 30 30 30 3 42.511 Indian Overseas Bank Public 10 15.5 20 1.8 0
42.512 Oriental Bank of Commerce Public 17.5 17.5 22.5 2.2513 Punjab National Bank Public 35 35 35 3.5 10014 Punjab & Sind Bank Public 25 25 25 3 015 State Bank of India Public 30 30 30 2.5 100
27016 Syndicate Bank Public 22.5 35 35 2.2517 UCO Bank Public 37.5 37.5 37.5 2.88 10018 Union Bank of India Public 21.88 21.88 43.75 3.85 55.6319 United Bank of India Public 19.5 33.5 42.5 2.25 5020 Vijaya Bank Public 24.5 34 34 2.25 3921 Axis Bank Limited Private 50 50 50 2.5 5522 Catholic Syrian Bank Private 12.5 20 30 2.25 5023 Dhanlakshmi-B Private 12.5 20 25 1.25 1024 Federal Bank Ltd. Private 25 25 25 2.5 87.5
![Page 8: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/8.jpg)
BASIC COMPUTATION
1/ 2
n m
i
1
i=1,…,mj =1,…,nS
w2
(x)2 x2 u /i ij j ijj
j 1
1/ 2
n m
i 1
i=1,…,mS w2 ( x )2 x2 v /i j ij j ij j =1,…,n j
1 Si C
i=1,…,mi Si Si
# Ranking according to increasing order of Ci
![Page 9: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/9.jpg)
TOPSIS RANK TABLE
S.No Bank name CategoryComposite Index(5 var)34
232811
Citibank N.A. Foreign 0.035992Dhanlakshmi-B Private 0.046986Karnataka Bank Ltd. Private 0.050524Indian Overseas Bank Public 0.064452
30 The Bank of Rajasthan Ltd.
Private 0.07398431122220
The South Indian Bank Ltd.
Private 0.082675Oriental Bank of Commerce
Public 0.086082Catholic Syrian Bank Private 0.088519Vijaya Bank Public 0.089792
19 United Bank of India Public 0.0933772 Andhra Bank Public 0.102432
14 Punjab & Sind Bank Public 0.10589632 ABN-AMRO Bank N.V. Foreign 0.10723229 Kotak Mahindra Bank Private 0.10934624 Federal Bank Ltd. Private 0.1095
725 HDFC Bank Ltd. Private 0.11258910 Indian Bank Public 0.11274521 Axis Bank Limited Private 0.11494633 Abu Dhabi Commercial
BankForeign 0.1152
715 State Bank of India Public 0.116578 Corp. Bank Public 0.118078
3 Bank of Baroda Public 0.120726
![Page 10: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/10.jpg)
KEMENY’S METHOD
Kendall tau distance (or “bubble sort distance”) K(,) = # of pairs of candidates (a,b) on which
and mismatchEx: K( (a b c d), (a d c b)) = 0 + 2 + 1 = 3
Kemeny Optimal aggregation[Kemeny 1959] Optimal aggregation w.r.t. Kendall tau distance
For a collection of given ordering τ1, τ2,…, τn on candidate a Kemeny optimal ordering minimizes the sum of the kendall tau distance
k
K , i i 1
![Page 11: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/11.jpg)
APPLICATION PROCEDURE:
We apply KEMENY’S method in two different ways:
Direct application of KEMENY’S method
Stepwise application of KEMENY’S method
![Page 12: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/12.jpg)
DIRECT APPLICATION OF KEMENY’S METHOD
Step 1:o we first decide number of candidate for applying
KEMENY.o we choose m=4(Banks A=1, B=2,C=3, D=4). We work out the 4! = 24 permutations of 1,2,3,4. A permutation2314 means that the ranking is BCAD.
Step 2:o we take our data matrix as 4 rows and n=6 columnso Construct rank position matrix with respect to each
of the 6 columns.
![Page 13: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/13.jpg)
DIRECT APPLICATION(CONT.)
Step 3:o we calculate the number of non matching pair
between each of the permutations of 1,2,3,4 &position ordering for 9 variable.
Step 4:o next calculate Kendall-distance for permutations.
o Sum kendall distance over permutation. And find permutation corresponding minimum kendall sum. This will be our ranked candidate.
![Page 14: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/14.jpg)
EXAMPLE
Position Rank Matrix:2 2 3 1 1 13 3 4 3 2 21 1 2 4 3 34 4 1 2 4 4
A=1
Indian Overseas Bank
0.06445
0.069847
0.094086
0.092438
13.5 12B=
2Dhanlakshmi-B 0.04
6990.050308
0.070184
0.226337
17.5 14.5C=
3Karnataka Bank Ltd.
0.05052
0.05131
0.126883
0.118069
27
22D=
4Oriental Bank of Commerce
0.08608
0.086876
0.198224
0.198181
34
25.5
![Page 15: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/15.jpg)
EXAMPLE(CONT.)
1 2 3 4
21
13
17
25
15
#Ranking is 1 2 3 4 (* min kendall sum)
Tranpose Position Matrix2 3 1 42 3 1 43 4 2 11 3 4 2
1 2 3 4
Kendall Tau Distance Kendall sum3 3 2 5 3 32 2 3 6 4 43 3 4 5 5 54 4 5 4 6 64 4 3 4 4 45 5 4 3 5 53 3 2 5 3 32 2 5 4 4 41 1 4 3 3 32 2 3 2 2 24 4 5 2 4 43 3 4 1 3 35 5 2 3 3 36 6 3 2 4 45 5 4 1 3 34 4 3 0 2 24 4 1 2 2 23 3 2 1 1 12 2 1 4 2 21 1 2 5 3 30 0 3 4 2 21 1 2 3 1 13 3 0 3 1 12 2 1 2 0 0
19212529232719
15
23
21
21
15111315119
117*
S. No. Permutation123456789101112131415161718192021222324
2 3 4 13 2 4 13 4 2 14 3 2 12 4 3 14 2 3 14 3 1 23 4 1 23 1 4 21 3 4 24 1 3 21 4 3 22 4 1 34 2 1 34 1 2 31 4 2 32 1 4 31 2 4 32 3 1 43 2 1 43 1 2 41 3 2 42 1 3 41 2 3 4
![Page 16: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/16.jpg)
STEPWISE OF KEMENY’SAPPLICATION METHOD
Example:
Step1: We, first fix the no. of rows(=k) to be taken at a time(say, 4). After fixing, we proceed as direct application.
Step2: The optimal Kemeny, thus obtained, is used toidentify the Rank 1 bank. In the next step, we remove rank
5th1 and add the(n-k+1).
bank then apply step 1. and go upto
A CitibankN.A. 0.03599
0.03346
0.329862
0.357041
51 45B IndianOverseasBank 0.0644
50.06985
0.094086
0.092438
13.5 12C Dhanlakshmi-B 0.0469
90.05031
0.070184
0.226337
17.5 14.5D KarnatakaBankLtd. 0.0505
20.05131
0.126883
0.118069
27 22E OrientalBankofCommerc
e0.08608
0.08688
0.198224
0.198181
34 25.5F CatholicSyrianBank 0.0885
20.0874 0.24697 0.21870
945 32
G TheSouthIndianBankLtd. 0.08268
0.08043
0.294886
0.280081
56 43
![Page 17: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/17.jpg)
STEPWISE APPLICATION(CONT.)
Step3:Repeat step1 & step2 the process tillranking stabilize.
Example(cont.): Here, four Kemeny’s steps in each iteration. After four iteration, rank of banks is stabilized.
1st iteration:
i1st Step 2nd Step 3rd Step 4th Step
SelectRow Rank
SelectRow Rank
SelectRow Rank
SelectRow RankBank
A A D . . . . . .B B A A E . . . .C C B B A A F . .D D C C B B A A AE . . E C C B B BF . . . . F C C CG . . . . . . G G
![Page 18: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/18.jpg)
STEPWISE APPLICATION(CONT.)2nd iteration:
3rd iteration:
iii1st Step 2nd Step 3rd Step 4th Step
SelectRow Rank
SelectRow Rank
SelectRow Rank
SelectRow RankBank
D D F . . . . . .F F D D C . . . .C C C C D D D . .G G G G G G G G BE . . E E E E E EA . . . . A A A AB . . . . . . B G
ii1st Step 2nd Step 3rd Step 4th Step
SelectRow Rank
SelectRow Rank
SelectRow Rank
SelectRow RankBank
D D D . . . . . .E E E E F . . . .F F F F A A C . .A A A A B B A A GB . . B E E E E EC . . . . C B B AG . . . . . . G B
![Page 19: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/19.jpg)
STEPWISE APPLICATION(CONT.)4th iteration:
4thOptimum combination(result) for iteration:B C D A E
,… iteratio
n.
F G5th 6thWe get same result after ,
iv1st Step 2nd Step 3rd Step 4th Step
SelectRow Rank
SelectRow Rank
SelectRow Rank
SelectRow RankBank
F F B . . . . . .C C F F C . . . .D D C C F F D . .B B D D D D F F AE . . E E E A A EA . . . . A E E FG . . . . . . G G
![Page 20: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/20.jpg)
STEPWISE APPLICATION(CONT.)
Ranking By Kemeny’s method:
# Here Indianthis method.
Overseas Bank is best according to
Kemeny's MethodBanks Categor
yB IndianOverseasBank Public
C Dhanlakshmi-B PrivateD KarnatakaBankLtd. PrivateA CitibankN.A. Foreign
E OrientalBankofCommerce
PrivateF CatholicSyrianBank PrivateG TheSouthIndianBankLtd. Public
![Page 21: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/21.jpg)
APPLICATION:
Meta search: Combine results of different web searchengines into a better over all rankingRank items in a database according to multiple criteriaOverall ranking of educational institute
o
ranking of Banks by no of customer, no of branches,o
different facilities etc.pollution ranking of citieslevel of poisonous gas like
on the basis of atmosphericCO2 ,SO2, NO2 etc.
o
Ex: Choose of Flight ticket by price, # of stops, date &time.Choosing car by style, reliability, fuel economy, cost.
o
o
![Page 22: Multi dimensional Ranking](https://reader035.vdocuments.us/reader035/viewer/2022081605/58ecd82d1a28abf31d8b45ef/html5/thumbnails/22.jpg)
.
Thank You