Вы находитесь на странице: 1из 25

Presentation on :- 1

Rank Prediction Using Feature Selection


and TransformationTechniques

GROUP : IT15

Presented By :
Prashant Pandey (20138055)
Jitendra Singh (20138044)
Ashish Kumar Yadav (20138051)
Under the Guidance of : Ashutosh Jha (20138005)
Er. Rupesh Kumar Dewang Vikas Babu Gond (20138064)
(Asst. Professor, MNNIT- Alld)
2
INTRODUCTION

• Respective position of Objects matters a lot.

• Ranks shows respective Development, Virtues and Qualities.

• We are very much interested to know our standings in a


particular area and improving it.

Vikas Babu Gond (20138064)


3
Cont..

• Comparison of attributes : The Only way of deciding


Ranks.

• What if number of attributes is too Large.

• Analysis becomes much complicated.

Vikas Babu Gond (20138064)


4
Cont…

• Reduction of attributes : a very useful option.

• Two available methods are feature selection


and feature transformation.

• Reduction of Analysis as well as cost and time overhead.

Vikas Babu Gond (20138064)


5
Control Flow Diagram start

Input data

Data Cleaning And


Processing
Applying algo of
feature selection and
give score to each
attributes.
Sort the score with
feature in decreasing
order
attributes which plays a
major role in determine
object
Calculation (i.e. score
and original values)

Sort the object based on


this calculation rating

stop
Vikas Babu Gond (20138064)
6
Data Description

• Country Dataset : 87 countries each having 43 attributes


namely total air craft, air craft carrier, nuclear weapons,
helicopters etc.

• IPL Dataset : 100 batsmen each having 6 attributes namely


innings, runs,average, strike rate, 4s, 6s.

Vikas Babu Gond (20138064)


7

PCA : Principal Component Analysis

• Simple feature Transformation technique.

• Works on co-relation between attributes of two objects.

• Aims to retain maximum information

Ashish Kumar Yadav (20138051)


8
Cont...

• Calculation of co-varience matrix and Eigen decomposition.

• Selecting Principal component.

• Deciding ranks on the basis of these Principal components.

Ashish Kumar Yadav (20138051)


9

Cumulative Graphs

Cumulative Graph of country Dataset using PCA Cumulative Graph of IPL Dataset using PCA

Ashish Kumar Yadav (20138051)


Cont… 10

Comparision of countries :

Standard Ranks of Countries Ranks of Countries decided by PCA

Ashish Kumar Yadav (20138051)


11
Cont...
Comparison of batsman rankings :

Standard Ranks of IPL Batsmen Ranks of IPL Batsmen decided by PCA


Ashish Kumar Yadav (20138051)
12
Random Forest

• Forest as the name suggest it is collection of Large


number of decision trees.

• Each tree is correct but they are missing a lot of data.

• Trees are highly different from each other.

Prashant Pandey (20138055)


13
Cont…

• Selection of random subset of variables from given all set


of variables.

• Making Decision trees for each subset.

• Taking output of all decision trees and averaging them to


to find final result.

Prashant Pandey (20138055)


14
Cumulative Graphs

Cumulative Graph of country Dataset using RF Cumulative Graph of IPL Dataset using RF

Prashant Pandey (20138055)


15
Cont…
Comparison of countries :

Standard Ranks of Countries Ranks of Countries decided by RF


Prashant Pandey (20138055)
16
Cont...
Comparison of batsman rankings :

Standard Ranks of IPL Batsmen Ranks of IPL Batsmen decided by PCA

Prashant Pandey (20138055)


17
Univariate Statistical Test

• Selection of a subset of relevant feature for model


construction.

• Automatically selects those features that contribute


the most.

• Includes or excludes attributes without changing them.

Jitendra singh (20138044)


18
Cont…

• Uses Chi-square Method.

• Shows the relationship between categorical variable.

• Chi- squared statistic : a number that shows the difference


between observed counts and the counts
expected if there were no relationship at all in the
population.

Jitendra singh (20138044)


19

Cumulative Graphs

Cumulative Graph of country Dataset using UST Cumulative Graph of IPL Dataset using UST

Jitendra singh (20138044)


20
Cont…
Comparison of countries :

Standard Ranks of Countries Ranks of Countries decided by UST


Jitendra singh (20138044)
21
Cont…
Comparison of batsman rankings :

Standard Ranks of IPL Batsmen Ranks of IPL Batsmen decided by UST

Jitendra singh (20138044)


22
Result and conclusion

• Which is better : Feature Transformation (PCA) or


Feature selection (Random forest , Univariate statistical
Test).

• PCA is better choice for deciding ranks.

• PCA consider all the attributes equally while feature


selection is dependent more on particular attributes.

Ashutosh Jha (20138005)


23
Cont…

• Feature selection can be improve by scaling up or down


attribute values according to their relevance.

• These techniques are purely statistical and based on


mathematical operations on data.

Ashutosh Jha (20138005)


24
References :

1. https ://wikipedia:org=wiki=Ranking
2. http : ==ww2:amstat:org=publications=jse=scariano:pdf.
3. http://==sebastianraschka:com=Articles=2014 pca step by
step:html.
4. http : ==www:globalfirepower:com=.

Ashutosh Jha (20138005)


25

Ashutosh Jha (20138005)

Вам также может понравиться