• No results found

Unsupervised feature selection using feature similarity

N/A
N/A
Protected

Academic year: 2023

Share "Unsupervised feature selection using feature similarity"

Copied!
12
0
0

Loading.... (view fulltext now)

Full text

(1)
(2)

8 "

B ' % $

$ #' C D ' #

% E 1 2 % # $ E2 ! 1' 2 +2

&

&

' ' $ # / 4/@ "

B # 9 $ # $

% F $ & - & 3

$

$ 7 1 ! 7

% 6 # # ++

8 1

2 % ! 4 ++

$ ++

$ 7 1

! 7

1112 % #!

# "

x y …x; y† ˆ cov…x;y†



var…x†var…y†

p ;

1 2 1 2 x y $ $ …x; y†1 1 x y …x; y† 0 81 j…x; y†j

x y %

F 1. 01 j…x; y†j 1

2. 1 j…x; y†j ˆ0 x y

3. 1 j…x; y†j ˆ1 j…y; x†j12

4. uˆx acy bd a; b; c; d 1 j…x; y†j ˆ1 j…u; v†j ( 5. % (

…x; y†

% / 3 ! &

&#

4 8 0 1e2 x y y yˆa‡bx a b '

e…x; y†2ˆ1 n

X…e…x; y†i†2;

e…x; y†iˆyi a bxi % aˆy bˆ…x;y†covvar…x† ' e…x; y† e…x; y† ˆvar…y†…1 …x; y†2† y x

e…x; y† ˆ0 x y

e…x; y† ˆvar…y† % e2 !

(3)

( y$ & e F

1. 0e…x; y† var…y†

2. e…x; y† ˆ0 x y 3. e…x; y† 6ˆe…y; x†12

4. uˆx=c vˆy=d a; b; c; d e…x; y† ˆd2e…u; v† e ( e (

5. % e ( x y

e 1 -2 : 1 32

7 1 ! 7122 = $ x y ( 7 1 ! 7 2…x; y† ˆ

22…x; y† ˆ …var…x† ‡var…y†



…var…x† ‡var…y††2 4var…x†var…y†…1 …x; y†2†

q :

% 2 2

1x; y2 + $ 1 2 ) % 1 2 ' 82 ) 1 2 % 1 1 7 1 1 !

% 2 $ #

++ 2 ' ' …x; y†

yˆa^‡bx ^

' % a^ˆxcot ‡y b^ˆ cot

ˆ2 tan 1 2cov…x; y† var…x†2 var…y†2

! :

% #' ? 2 F

1. 02…x; y† 0:5…var…x† ‡var…y††

2. 2…x; y† ˆ0 x y 3. 2…x; y† ˆ2…y; x†12

4. uˆxcyd a; b; c; d 2…x; y† 6ˆ2…u; v† ( & $ 2

(

5. 2 ( 1 2 $2

% 2 ! 1 -2 1 /2 1 32 1x; y2 $ 8

% ! 12 # k-NN & + k

$* )$ -< +)*1 )( )$*) )-)$ ,+ +0 )$*) -* $. 454

6 + " < => -&' " =e> => &' ? " =2>

(4)

1 2 k %

k # 12 ' kth # ' ! 2 k

%k % ? + % F Algorithm:

= D Oˆ fFi; iˆ1;. . .; Dg 0 # Fi Fj S…Fi; Fj† 8 S % 1 ; e; 22

& + S =rki

Fi kth # R %

Step 1F " kD 1

R O R O

Step 2F ? Fi2R rki Step 3F ? Fi0 rki0

0 R k Fi0 1FFi0 k # R2Letˆrki0

Step 4F Ifk >cardinality…R† 1Fkˆcardinality…R† 1 Step 5F Ifkˆ1FGo to Step 8

Step 6F Whilerki0> doF 1 2kˆk 1

rki0ˆinfFi2Rrki

1CkD 1

Ckth #D

R -dissimilar

2 12Ifkˆ1FGo to Step 8

1 R -dissimilar

C #D R2

End While Step 7FGo to Step 2

Step 8F 0 R

Remarks:

! !7$ % # # $ 1D2 $ O…D2† $ ' ! $ O…D2† 6 ! #l# !#r ' + $ ' : ' '

% # # $ l $ O…l† % $ O…D2l† 1 K-NN # 2 O…l2† $ : #

5 1 3 1k k & k 12 % % k

5 5 1 $ % G! # ! + #

% *# . $ 2

(5)

. F = l c D O d R O D R

d

1. !"$+ " S Sˆtrace…Sb1Sw†Sw $ Sb $ F

Sw ˆ Xc

1

jEf…X j†…X j†Tj!jg ˆXc

1

jj

Sb ˆ Xc

1

…j Mo†…j Mo†T

Moˆ EfXg ˆXc

jˆ1

jj;

…1† j !jX j

!j Mo j

$ !j Ef:g

$ # S

2. 955 1 +$ 8

*#

"#

E A @A % # % * *# '

3. 5( 2$ 1 +$ . $

! + : A @A %

4. !$ A = p; q

Dpqˆ XM

jˆ1

xp;j xq;j

maxj minj

2

" #1=2

; xp;j p jth maxj; minj $ jth $ M &

pq sim…p; q† ˆe Dpq

$* )$ -< +)*1 )( )$*) )-)$ ,+ +0 )$*) -* $. 45@

$A-) 6

" " -&( (

E: Entropy, FFEI: Fuzzy Feature Evaluation Index, S: Class Separability, KNNA:k-NNclassification accuracy, BayesA: naive Bayes classification accuracy, and SD: standard deviation. SFS: Sequential Forward Search and SWC: Stepwise Clustering. d: number of selected features, D: number of original features andk: parameter used by the proposed method.

(6)

ln0:5D D 9 F

Eˆ Xl

1

Xl

1

…sim…p; q† log sim…p;q†

‡ …1 sim…p; q†† log…1 sim…p;q†††: …2†

$ # 5. 3::$ 3 ( 7 - ?

$ 1??9 2 F F F EIˆ 2

l…l 1† X

p

X

q6ˆp

1 2 Rpq…1 Opq† ‡Opq…1 Rpq†

h i

;

…3†

Opq Rpq p q O R : # pq

pq ˆ1 Ddpq

max ifdpq Dmax

ˆ0; otherwise:

dpq p q Dmax $

% ??9 B B 8 ??9 *# # # $ = '

$

6. 0! !$+ = dd $ d j; jˆ1;. . .; d =

~jˆ j

Pd 1j

:

$A-) 2

" " &( (

BB: Branch and Bound, SFFS: Sequential Floating Forward Search.

(7)

~j ! 0~j1 Pd

jˆ1jˆ1 8

HRˆ Xd

1

~j log ~j: …4†

% HR 12 $ ' ' HR $ % ! ! !$

! "$ ! 1 1 % ' # &

1 2 $

d-dimensional

D-dimensional

* #= + 1 d 2 8 # *#

= *#=

$

!

"#

# 6 $ F ? %

& / $ ' # 9 k

$* )$ -< +)*1 )( )$*) )-)$ ,+ +0 )$*) -* $. 45;

$A-) 4

" " -!&( (

(8)

% # F # 1D102 # 110< D1002 # 1D >1002 # % G" : = 0 +- %

1. % 9 3A )

% 47 ‰0;1Š 77@7 +4

2. ! 3 % 1CAD#C@D2 $ ( % +AAA 4/@ A

3. +$ /3+

+7@ @3 $ % 9", # ! 4

4. !" % ! % /4A 37 # ' +

5. ;(1 % 3AAA /A % % !

6. ! % # % ! # % -3 -/

7. 3 ( $! % , &

% 3/

A

% 3<A+

8. ; % 4</

9. % 3A

!$ %&'% ( )'*''% )+' ,

#*%

? 1+2

$ 1-2 12 *# . ? F

1. . . 1..2 + $ ..

2. &' ? & 1&?&2 + $

3. &' ? ? & 1&??&2 - # .. ' ! !

4 1 " ! # "

"=> " => =>

$ " ! $ 6 2 4

(9)

4. & " 1 2 1&"2 +

$ 1+2

"

% + - # ! $ "

? - 12 12 12 % "G ' & G & -3A :8 ! % + - &

1..2 ' 1&??&2 ' % ? 1 *#

. 2 1&(2

" # 1.. &??&

&?&2 # ' 6 &"

! ' ? $ #

.. &??& -A#3A ? # AA # ..

&??& AA &?&

# &?&

AA .. &??&

' 1 .. &??& &?&2 CD * % + - 8 $

$ 1??9 2 1-2 % / 1 : ? ? "2 % #

$ # 0#?

3A 0#? 0 % + - % 0#? # : ' ?

&

: &(

A t-test ' % .#? #

$* )$ -< +)*1 )( )$*) )-)$ ,+ +0 )$*) -* $. 45B

$A-) 3

" " - ( ! )

(10)

+/ +3 &" &?&

? 0#? .. &??& #

!$ + - +'% ( .+ ''/ +-

0 1HR2 & / HR

1 2 1 2

= HR HgR HR

HRs

% 3 HRs HRs

% 7 1

! 7 2 1 e2

% 4 HRs HRg

$A-) 7

* * ( ""

HRg: Average representation entropy of feature groups,HsR: representation entropy of selected subset,2: maximal information compression index, e: least-square regression error, and: correlation coefficients.

$A-) @

* )HRs "

% vˆ x1 x2

1s21‡2s22

p x1;x2 s1; s2 1ˆ1=n1; 2ˆ1=n2n1; n2

(11)

% 4 2

HgR HRs

!$ ** %* #k

k ? / E $ k 8 # 1? / 2 k k k #

k ?

d‡kD d D

0

% # ' 6 ' "G G!

$ $ % ! ! B % k ! ! 1 2

. 7 1 ! 7 6 ! % # $

$ $ ! !$

$ ' #

%

B 6 # ++

# 5 $ ! 1 2

$* )$ -< +)*1 )( )$*) )-)$ ,+ +0 )$*) -* $. 466

3 1 # " ! k

"=> " => =>

(12)

G ? 0 G C( : * ( ( D + -@ +/#+7 @@4

[2] () H * < 0 + +!!9 "F 8 @<+

[3] H I H * C? & :

? &D< 0 4 3 @#

+3 @@/

[4] ( 0= . ! C " 9

&' ? & D +1 =( ? H#8 = J!F &

> @@4

[5] > + 1 < 0&*

. 0 F "0" @@4

[6] : * H &! ! C" &

? " D< 0 -- +3#

/ +AAA

[7] ( &! ! C ? & &

0 : 8 " D < )) ? 4 1 +@-#-A @@/

[8] : :& = C9 :

" > 9D< )) ? 1 4

@@/

[9] 8 = 0 & C& & ? &D 7! $ +!! 3 ---#--@ @@<

[10] : ( 8 = CG ? &D <

<1 + 1 9 @ ($ @ A#+

+AAA

[11] H ( " . C? & & 6 G = D< ) ? 1 4+AAA

[12] & . " : 6 C: $ 9 : $ =! " ? & :#

( D< ? $! $ # +47#+7A +AAA

[13] &* 0* ( H . ! CG ? 9 F

#? D 5 5%

-44#-74 +AAA

[14] : 8 C" . ? & ( " : = D< ) ? 1 4+AAA

[15] 0 8 C0 ? 9$ D

! A3#A3/ @7

[16] &* ( C? & = ( : D ! A4#A@ @7

[17] ,% % %0 > C" ?

& = ( : D

! /A< @7+

[18] * * = 0 C ?

&D< 5 ? ;% ! 4 +/@#

+34 @@+

[19] *! C9 F 9$

0D< ( ! 4 1 7#<+

@@/

[20] ( * : & C% 6 ? &D

< ) ? 1 4 +</#+@+ @@4 [21] . * C&# " D A +

+ <4#A @47

[22] "0 0 4 1 +!! H

@7-

[23] "=. ! "H:0! $14@"

G " ( "

& FBBB B:=0

@@<

[24] 9= = 1 B$! J!F H @74

[25] C% G "

% > D2% -4 +/3#+7 @/@

#' ' A$

" "

$ 6BB7 / ! ! " * "!

/ " ! &

"

/ " )))

$$ +1- A =/>

( "

= > / &

) - 6BB6&B2 6BB7&B; / "

" /

" "

! " ! / ! 6BB7 " " ) / "! "

+ " )

2 3$ #)= C86& C83& CB4> =53245474>

$ ( 6B;3 6B;B " "

6B82 ( ! ( "

" - /

"

/ " "

/ ! " " A "

6B87&8;

1 ! + D / $%

6BB5&B2 6BB3 +*&

+ * E / / 6BBB " / " ))) =>

" & " * 6BB;&BB / "! " )))

$ F " " + " 9) / "

"## / 9 " Fuzzy Mathematical Approach to Pattern Recognition D F =/> +! . 6B87Neuro-Fuzzy Pattern Recognition: Methods in Soft Computing D F +! . 6BBB 455 / 6BB5 A # =!

! " > 6BB4 D! +

! 6BB4 1 * ! 6BB4 + $ A " ! 6BB3 ))) $ + +! , ! 6BB@ + ! 6BB; )$)&*

- F! 0 6BB8 , A ! 6BBB 0( A ! " " * 2555 ! # &

! =6 ! > " * " 2556 / G " + 2556 ! ) $ "

"IEEE Transactions on Neural Networks

=6BB3&B8> Pattern Recognition Letters International Journal Pattern Recognition and Artificial IntelligenceNeurocomputing, Applied Intelli- gence Information Sciences, Fuzzy Sets and Systems Funda- menta Informaticae / " )% ) A IEEE Transactions Fuzzy Systems International Journal on Image and Graphics International Journal of Approximate Reason-

ing " ? IEEE Computer

. % % ' *%'% % 1' % - %1 %&+' , %&'4

&) /'' %+ ',') '- <999 9

References

Related documents

National Institute Of Technology Rourkela Page 52 Fig 5.11 (a) variation of Nu with respect to Re (b) variation of pumping power with respect to Re (c) variation of friction

After localization of the iris, Scale Invariant Feature Transform (SIFT) is used to extract the local features.. The SIFT descriptor is a widely used method for matching

Hence to supplement the complimentary features of the SIFT and SURF, a new Feature based image mosaicing technique using image fusion has been proposed and

This chapter is based on the details of the face recognition method using swarm optimization based selected features.. Chapter 6: Results

Hence a novel feature selection method has also been proposed which uses Denoising Autoencoder with correlation based multiplicative aggregation function to select relevant

105 6.5 Feature recognition from segmented model using proposed algorithm 105 6.6 The Segmented mesh object representation using our data structure 107 6.7 Features and

tion 3, we put forward the concept of fuzzy-rough sets on compact computational domain. Based on this definition, Section 4 builds improved feature selection algorithm, which is

The candidates bearing the following Roll Numbers are declared to have passed the 1ST SEMESTER B.A.. KAKOTY) Controller of Examinations. Dibrugarh University