K-Means Clustering
K-Means Clustering
K-Means Clustering
3ULYDF\3UHVHUYLQJ.PHDQV&OXVWHULQJ$6XUYH\
5HVHDUFK
)DWLPD0HVNLQHDQG6DILD1DLW%DKORXO
'HSDUWPHQWRI&RPSXWHU6FLHQFH8QLYHUVLW\RI(V6HQLD$OJHULD
'HSDUWPHQWRI&RPSXWHU6FLHQFH8QLYHUVLW\2UDQ$OJHULD
$EVWUDFW7KHFOXVWHULQJLVDQH[SORUDWRU\WDVNRIGDWDPLQLQJ7KLVUDLVHGDUHDOSUREOHPRISULYDF\ZKHQWKHGDWDDUH
IURP GLIIHUHQW VRXUFHV 0RVW RI UHVHDUFKHV RQ SULYDF\ SUHVHUYDWLRQ LQ FOXVWHULQJ DUH GHYHORSHG IRU NPHDQV FOXVWHULQJ
DOJRULWKPE\DSSO\LQJWKHVHFXUHPXOWLSDUW\FRPSXWDWLRQIUDPHZRUN7KHGLVWULEXWLRQRIGDWDPD\EHGLIIHUHQWYHUWLFDO
KRUL]RQWDO RU DUELWUDU\ $SSURDFKHV DOORZLQJ VROYLQJ WKH SUREOHP RQ D YHUWLFDO KRUL]RQWDO DQG HYHQ DUELWUDU\
SDUWLWLRQHGGDWDVHWZHUHSURSRVHG7KHPDMRULQWHUHVWLVWRUHYHDOWKHPLQLPXPRILQIRUPDWLRQGXULQJWKHH[HFXWLRQRIWKH
DOJRULWKPHVSHFLDOO\LQNPHDQVLWHUDWLRQVZKLFKSRVHVDUHDOFKDOOHQJHIRUVHFXUHPXOWLSDUW\FRPSXWDWLRQ7KLVZRUN
FRQVLVWVWRVWXG\DQGDQDO\]HDOOZRUNVRISULYDF\SUHVHUYLQJLQWKHNPHDQVDOJRULWKPFODVVLI\WKHYDULRXVDSSURDFKHV
DFFRUGLQJWRWKHXVHGGDWDGLVWULEXWLRQZKLOHSUHVHQWLQJWKHZHDNQHVVHVDQGWKHVWURQJSRLQWVRIHDFKSURWRFROUHJDUGVWR
SULYDF\7KHLQWHUHVWLVWRDULVHWKHUHDOQHHGVRISULYDF\GXULQJWKHH[HFXWLRQRIWKHGLIIHUHQWVWHSVRINPHDQDOJRULWKP
WKXVWRGLVFRYHUWKHEHVWRIDSSURDFKHVLQFDVHRISUHVHUYLQJSULYDF\LQNPHDQVDOJRULWKP
.H\ZRUGVNPHDQVFOXVWHULQJDOJRULWKPVHFXUHPXOWLSDUW\FRPSXWDWLRQGLVWULEXWHGGDWDSULYDF\SUHVHUYLQJ
5HFHLYHG'HFHPEHUDFFHSWHG0DUFK
,QWURGXFWLRQ SUHVHUYLQJ GDWD PLQLQJ RQHV LV LPSRUWDQW GXH WR WKH
XVH RI WKLV DOJRULWKP LQ LPSRUWDQW RWKHU DUHDV OLNH
7KHYHU\DFWLYHUHVHDUFKDUHDRISULYDF\SUHVHUYLQJGDWD LPDJH DQG VLJQDO SURFHVVLQJ ZKHUH WKH SUREOHP RI
PLQLQJ DLPV WR H[WUDFW XVHIXO LQIRUPDWLRQ IURP GDWD VHFXULW\ LV VWURQJO\ SRVHG >@ 0RVW RI ZRUNV LQ
FRPLQJ IURP PXOWLSOH VRXUFHV ZKLOH SUHVHUYLQJ WKHVH SULYDF\SUHVHUYLQJFOXVWHULQJDUHGHYHORSHGRQWKHN
GDWD DJDLQVW GLVFORVXUH RU ORVV 7KH ILUVW ZRUNV RI PHDQV DOJRULWKP E\ DSSO\LQJ WKH PRGHO RI VHFXUH
SULYDF\ SUHVHUYLQJ GDWD PLQLQJ ZHUH JLYHQ E\ > @ PXOWL SDUW\ FRPSXWDWLRQ RQ GLIIHUHQW GDWD
RQ WKH ,' DOJRULWKP IRU GHFLVLRQ WUHHV FODVVLILFDWLRQ GLVWULEXWLRQV YHUWLFDOO\ KRUL]RQWDOO\ DQG DUELWUDU\
>@ (DFK ZRUN XVHG D GLIIHUHQW PRGHO RI SULYDF\ ,Q SDUWLWLRQHG GDWD ,Q WKLV ZRUN ZH EXLOG DQ RYHUYLHZ
>@ SULYDF\ LV SUHVHUYHG E\ UDQGRPL]LQJ WKH RULJLQDO RI SUHYLRXV ZRUNV LQ SULYDF\ SUHVHUYLQJ NPHDQV
GDWDVHW DQGWKHQ SURYLGHVLW DV LQSXWWR WKH DOJRULWKP FOXVWHULQJ DOJRULWKP EDVHG RQ VHFXUH PXOWLSDUW\
,Q >@ D PRUH ULJRURXV PRGHO LV XVHG ZKLFK LV D FRPSXWDWLRQ ZH FODVVLI\ WKHP DFFRUGLQJ WR GDWD
VHFXUH PXOWL SDUW\ FRPSXWDWLRQ > @ ,Q WKLV GLVWULEXWLRQ EHFDXVH LW LV WKH RQO\ SDUDPHWHU ZKLFK
DSSURDFKFU\SWRJUDSKLFSULPLWLYHVDUHDGGHGWRWKH,' DIIHFWVWKHDSSURDFKWREHXQGHUWDNHQ:HSUHVHQWWKH
DOJRULWKP LQ RUGHU WR SURWHFW GLVWULEXWHG GDWD IURP ZHDNQHVVHV DQG VWUHQJWKV RI HDFK SURSRVHG VROXWLRQ
GLVFORVXUH ZKHQ H[HFXWLQJ LW 6LQFH PRUH UHVHDUFKHV LQ WHUPV RI SULYDF\SUHVHUYLQJ DQG FRPSXWLQJ FRVW
EDVHGRQVHFXUHPXOWLSDUW\FRPSXWDWLRQIRUGLIIHUHQW ZKLFK DUH DPRQJ PHWULFV WR HYDOXDWH DQ\ SULYDF\
DOJRULWKPVRIGDWDPLQLQJZHUHHVWDEOLVKHG>@ SUHVHUYLQJGDWDPLQLQJDOJRULWKP>@
/DWHO\ PDQ\ VWXGLHV ZHUH LQWHUHVWHG E\ YLHZV DQG
VXUYH\V RQ SULYDF\ SUHVHUYLQJ GDWD PLQLQJ ZRUNV >
@ LQ DWWHPSW DW JHQHUDOL]DWLRQ WKH UHVXOWV IRU .PHDQV&OXVWHULQJ$OJRULWKP
GLIIHUHQW FDWHJRULHV RI GDWD PLQLQJ PHWKRGV :H $PRQJ WKH IRUPXODWLRQV RI SDUWLWLRQDO FOXVWHULQJ
EHOLHYH WKDW WKH GLYHUVLW\ RI GDWD PLQLQJ PHWKRGV DQG EDVHGRQWKHPLQLPL]DWLRQRIDQREMHFWLYHIXQFWLRQN
WKHLUVSHFLILFXVHOLPLWVWKHDSSOLHGSULYDF\SUHVHUYLQJ PHDQV DOJRULWKP >@ LV WKH PRVW ZLGHO\ XVHG DQG
PRGHOVWRWKHVHPHWKRGV7REHWWHUFRQWUROWKLVDUHDLW VWXGLHG*LYHQDGDWDVHW'RIQHQWLWLHVREMHFWVGDWD
LVEHVWWRLVRODWHHDFKRQHDQGVHSDUDWHO\VWXG\LW SRLQWVLWHPV LQ UHDO SGLPHQVLRQ VSDFH 5S DQGDQ
&OXVWHULQJ>@LVDGDWDPLQLQJPHWKRGWKDWKDVQRW LQWHJHUN7KH.PHDQVFOXVWHULQJDOJRULWKPSDUWLWLRQV
WDNHQLWVUHDOSDUWLQWKHZRUNVDOUHDG\TXRWHGDOWKRXJK WKHGDWDVHW'RIHQWLWLHVLQWRNGLVMRLQWVXEVHWVFDOOHG
WKH PRVW LPSRUWDQW DOJRULWKP RI WKLV PHWKRG ZDV YHU\ FOXVWHUV (DFK FOXVWHU LV UHSUHVHQWHG E\ LWV FHQWHU
VWXGLHGLQWKHFRQWH[WRISULYDF\SUHVHUYLQJZKLFKLVN ZKLFK LV WKH FHQWURLG RI DOO HQWLWLHV LQ WKDW VXEVHW
PHDQVDOJRULWKP>@6XUYH\LQJSULYDF\SUHVHUYLQJN
PHDQV FOXVWHULQJ DSSURDFKHV DSDUW IURP RWKHU SULYDF\
3ULYDF\3UHVHUYLQJ.PHDQV&OXVWHULQJ$6XUYH\5HVHDUFK
SDUWLHVRQO\,QWKHGHVFULEHGVFHQDULRRQO\WKHHQWLWLHV DQGDVHFXUHDSSUR[LPDWLRQWHFKQLTXH>@+RZHYHU
RI HDFK SDUW\ DUH NHSW FRQILGHQWLDO WKH LQWHUPHGLDWH WKH SURSRVHG VROXWLRQ GRHV QRW VROYH WKH SUREOHP RI
FHQWHUVDUHUHYHDOHGWRWKHWZRSDUWLHVDQGWKHGLVWDQFH WKHORFDOGLYLVLRQ ZKLFKOHDNV WKH QXPEHU RI HQWLWLHV
FRPSXWDWLRQ LV GRQH ORFDOO\ LQ HDFK SDUW\ 6HFXULW\ LQ HDFK FOXVWHU WR GLIIHUHQW SDUWLHV GXULQJ FHQWHUV
SULPLWLYHVDUHXVHGIRUFHQWHUVFRPSXWDWLRQLQRUGHUWR FRPSXWDWLRQ DQG WKH UHYHODWLRQ RI WKH LQWHUPHGLDWH
SUHVHUYHWKHSULYDF\RIHQWLWLHVRIHDFKSDUW\7KHILUVW DVVLJQPHQWVRIHQWLWLHVWRFOXVWHUV
SURWRFROFDOOHG23(LVEDVHGRQREOLYLRXVSRO\QRPLDO %XQQ DQG 2VWURYVN\ >@ RIIHU D PRUH HIILFLHQW
HYDOXDWLRQJLYHQE\1DRUDQG3LQNDV>@7KHVHFRQG VROXWLRQ LQ SULYDF\ IRU WKH NPHDQV DOJRULWKP RQ
SURWRFRO QDPHG '3( LV EDVHG RQ KRPRPRUSKLF DUELWUDULO\ SDUWLWLRQHG GDWD 7KH DXWKRUV SURYLGH D
HQFU\SWLRQVFKHPHV ULJRURXV DSSURDFK WR VHFXULW\ LQ WKH VHPLKRQHVW
7KHSURSRVHGVROXWLRQVWLOOUHTXLUHVWKDWERWKSDUWLHV PRGHOWKHLUSURWRFROGRHVQRWUHYHDOWKHLQWHUPHGLDWH
DUH VHPL KRQHVW WKH DLP ZDV WR H[SHULPHQWDOO\ FHQWHUVDQGFOXVWHUVDVVLJQPHQWVWKH\DOVRHIIHFWLYHO\
HYDOXDWH WKH WZR SURWRFROV 7KH KRPRPRUSKLF VROYHWKHSUREOHPRIWKHORFDOGLYLVLRQ7KH\GHYHORS
HQFU\SWLRQ VFKHPH LV PRUH HIILFLHQW WKDQ 23( IRU WKH VHFXUHVXESURWRFROVIRUHDFKRSHUDWLRQLQWKHNPHDQV
WZR SDUDPHWHUV FRPSXWLQJ DQG FRPPXQLFDWLRQ FRVW DOJRULWKP RQ WKH EDVLV RI 3DLOOLHU KRPRPRUSKLF
EXWWKHLUVROXWLRQFDQQRWEHH[WHQGHGWRVHYHUDOSDUWLHV FU\SWRV\VWHPV >@ HVSHFLDOO\ IRU VHFXUH PXOWLSDUW\
6DPHWHWDO>@KDYHSURSRVHGDSURWRFROZKLFKXVHVD GLYLVLRQ 7KH DXWKRUV SURSRVH DOVR D VHFXUH SURWRFRO
VHFXUH PHWKRG RI GLYLVLRQ WKDW SURWHFWV WKH HQWLWLHV RI IRU UDQGRP FKRLFH RI N LQLWLDO FHQWHUV 'HVSLWH WKLV
HDFKSDUW\DQGSUHYHQWVWKHUHYHODWLRQRIWKHQXPEHURI WKHLUVROXWLRQIDLOVWRPDNHFRQILGHQWLDOWKHQXPEHURI
HQWLWLHVLQHDFKFOXVWHU7KHSURWRFROLVDOVRDSSOLFDEOH LWHUDWLRQVLQWKHDOJRULWKP
LQ D PXOWLSDUW\ HQYLURQPHQW EXW WKH LQWHUPHGLDWH 6DNXPD DQG .RED\DVKL >@ JLYH D QHZ SURWRFRO
FHQWHUVDUHDOZD\VUHYHDOHG EXWDSSOLHGWRVHYHUDOSDUWLHVFDOOHGQRGHV7KHDLPLV
WRSURYLGHDSURWRFROIRUSUHVHUYLQJSULYDF\WRVLPSOH
$GYDQFHG:RUNVLQ3ULYDF\3UHVHUYLQJ. XVHUV UDWKHU WKDQ RUJDQL]DWLRQV RU FRPSDQLHV ,Q WKLV
0HDQVRQ$UELWUDULO\'LVWULEXWHG'DWD FDVHWKHSURWRFROLVVFDODEOHDFFRUGLQJWRWKHQXPEHU
RISDUWLHV7KHLGHDLVWRVHFXUHWKHSURWRFROSURSRVHG
$OWKRXJKDQDUELWUDULO\SDUWLWLRQHGGDWDVHWLVXQOLNHO\ E\.RZDOVF]\NDQG9ODVVLV>@ZKLFKFDOFXODWHVWKH
LQ SUDFWLFH WKH LQWHUHVW RI FRQVLGHULQJ VXFK D DYHUDJH RI WKH YDOXHV GLVWULEXWHG WKURXJK 33
SDUWLWLRQLQJ LV WKDW WKH SURWRFROV LQ WKLV PRGHO FDQ EH QHWZRUNV ZLWKRXW WUDQVIHUULQJ GDWD WR D FHQWUDO
DSSOLHG ERWK WR KRUL]RQWDOO\ SDUWLWLRQHG GDWD DQG GHSRVLWRU\ IRU FRPSXWLQJ FHQWHUV $W WKLV OHYHO
YHUWLFDOO\ SDUWLWLRQHG GDWD ,Q WKLV FDVH WKH QHHG RI SULYDF\ LV SUHVHUYHG XVLQJ 3DLOOLHU KRPRPRUSKLF
SULYDF\LQWKHNPHDQVDOJRULWKPLQFOXGHVDOOWKHFDVHV FU\SWR V\VWHPV )RU WKH SURWHFWLRQ RI GLVWDQFHV SDUWV
DOUHDG\VHHQ <DRHYDOXDWLRQFLUFXLWDQGWKHUDQGRPVKDUHVSULQFLSOH
7KH LGHD RI SULYDF\SUHVHUYLQJ SURWRFRO RQ DUHLQWURGXFHG7KHFDOFXODWLRQLVGRQHIRUHDFKQRGH
DUELWUDULO\ SDUWLWLRQHG GDWD ZDV LQWURGXFHG E\ ZKLFK PDNHV WKH SURWRFRO VFDODEOH DQG IDXOW WROHUDQW
-DJDQQDWKDQ DQG :ULJKW >@7KHLUVROXWLRQ SUHVHUYHV %XWWKHUHLVQRLPSURYHPHQWLQVHFXULW\
WKH SULYDF\ LQ WKH NPHDQV DOJRULWKP EHWZHHQ WZR
SDUWLHV1RRWKHUFRQILGHQFHSDUW\LVXVHG7KHLGHDLV &RQFOXVLRQV
WKDW DOO WKH FDOFXODWHG YDOXHV LQ WKH LQWHUPHGLDWH VWHSV
DUH VSOLW WR UDQGRP YDOXHV WKH SULQFLSOH RI UDQGRP $OWKRXJKWKHXVHGVHFXULW\PRGHOLVUHODWLYHO\VLPSOH
VKDUHV 7KH VHFXUH VFDODU SURGXFW >@ LV XVHG IRU VHPLKRQHVWPRGHOWKHSURWRFROVSURSRVHGGRQRW
VHFXUH FRPSXWDWLRQ RI GLVWDQFHV DQG WKH <DR FLUFXLW SURYLGHDFRPSOHWHSUHVHUYDWLRQRISULYDF\LQNPHDQV
HYDOXDWLRQ LV XVHG IRU VHFXUH FRPSDULVRQ DQG DOJRULWKPWKHSUREOHPLVLQWKHLWHUDWLYHQDWXUHRIWKH
FDOFXODWLRQ RI LQWHUPHGLDWH FHQWHUV 7KHVH VHFXULW\ DOJRULWKP %HVLGHV WKH LQIRUPDWLRQ WR SURWHFW DUH
SULPLWLYHV DUH H[HFXWHG IRU HDFK HQWLW\ RI WKH GDWDVHW QXPHURXVGLVWDQFHVYDOXHVLQWHUPHGLDWHDVVLJQPHQWV
ZKLFKUHIHUVWKHSURWRFROWRWKHVDPHSUREOHPRIZKLFK WR FOXVWHUV QXPEHU RI SRLQWV LQ HDFK FOXVWHU
SURSRVHG E\ 9DLG\D >@ WKDW RI D KLJK FRPSXWDWLRQ LQWHUPHGLDWH FHQWHUV QXPEHU RI LWHUDWLRQV DQG WKH
FRVWIRUDODUJHGDWDVHW HQWLWLHV YDOXHV WKHPVHOYHV 7KH PRGHO RI WKH GDWD
7KH VROXWLRQ KDV D ZHDNQHVV DW WKH XSGDWHG FHQWHUV GLVWULEXWLRQ DOVR DIIHFWV WKH ZD\ WKDW SULYDF\ LV
ZKHUHWKH\FRQVLGHUWKHGLYLVLRQDVDPXOWLSOLFDWLRQE\ SUHVHUYHG
WKHLQYHUVHZKLFKGRHVQRWLPSOHPHQWFRUUHFWO\WKHN $UELWUDULO\SDUWLWLRQLQJGDWDVHWLVEHVWVXLWHGWRVHHN
PHDQVDOJRULWKP$QRWKHUSURWRFROZDVJLYHQE\6XHW D PRUH JHQHUDO VROXWLRQ %XW PRVW ZRUN LQ WKLV
DO >@ ZKHUH WKH\ LQWURGXFH VHFXUH GDWD GLVWULEXWLRQ PRGHO LV DSSOLHG WR RQO\ WZR SDUWLHV
VWDQGDUGL]DWLRQLQRUGHUWRJLYHJUHDWHUDFFXUDF\WRWKH H[FHSW WKDW LQ >@ $OO ZRUN RQ WKH YHUWLFDOO\
UHVXOW RI FOXVWHULQJ $Q LPSURYHPHQW LQ VHFXULW\ SDUWLWLRQHGGDWDVHWPRGHOLVJLYHQRQPXOWLSOHSDUWLHV
FRPSDUHGWRWKHILUVWSURWRFROLVWKDWWKHFRPSDULVRQRI EXW FRQVLGHULQJ QRW FROOXGLQJ WUXVW SDUWLHV > @
GLVWDQFHVDQGFHQWHUVFRPSXWLQJDUHGRQHE\WKHVHFXUH ZKHUH QRVHFXULW\ JXDUDQWHHV LV JLYHQ LI WKHVH SDUWLHV
VFDODU SURGXFW VHFXUH REOLYLRXV SRO\QRPLDO HYDOXDWLRQ DJUHHG7KHFRVWRISULYDF\LVPHDVXUHGUHODWLYHWRWKH
7KH,QWHUQDWLRQDO$UDE-RXUQDORI,QIRUPDWLRQ7HFKQRORJ\9RO1R0DUFK
6,*02'5HFRUGYROQRSS
>@ <DR $ +RZ WR *HQHUDWH DQG ([FKDQJH
6HFUHWV LQ 3URFHHGLQJV RI WKH WK ,(((
6\PSRVLXPRQ)RXQGDWLRQVRI&RPSXWHU6FLHQFH
7RURQWR&DQDGDSS
)DWLPD0HVNLQHLVDGRFWRUDOVWXGHQW
DW WKH GHSDUWPHQW RI FRPSXWHU
VFLHQFH IDFXOW\ DW WKH 8QLYHUVLW\ RI
2UDQ DQG D PHPEHU RI LQIRUPDWLRQ
DQG LQIRUPDWLRQ WHFKQRORJ\
ODERUDWRU\RI2UDQ/,7,2DSSURYHG
LQ 6KH LV DOVR WKH HQJLQHHU RI
WKHFRPSXWHUVFLHQFHFHQWHUDW0RVWDJDQHP8QLYHULVW\
6DILD1DLW%DKORXOLVDOHFWXUHUDWWKH
&RPSXWHU 6FLHQFH 'HSDUWPHQW
6FLHQFH )DFXOW\ DW WKH 8QLYHUVLW\ RI
2UDQ (V6HQLD6KH REWDLQHG DIWHU
VHYHUDO VFLHQWLILF VWD\V LQ &1$0 RI
3DULV DQG /,5,6 /DERUDWRU\ WKH
8QLYHUVLW\ RI &ODXGH %HUQDUG
/\RQ KHU 3K' GHJUHH DW WKH 8QLYHUVLW\ RI 2UDQ LQ
6KHLVDPHPEHURI,QIRUPDWLRQDQG,QIRUPDWLRQ
7HFKQRORJ\ /DERUDWRU\ RI2UDQ /,7,2 ZKLFK LV
DSSURYHG LQ 6LQFH VKH KDV EHHQ OHDGLQJ D
WHDP RQ WKH7RSLF RI 'DWD (QJLQHHULQJ DQG :HE
7HFKQRORJ\+HU UHVHDUFKFRYHUV DGYDQFHG DVSHFWV RI
GDWDEDVHV ZHE WHFKQRORJ\ DQGXQVXSHUYLVHG
FODVVLILFDWLRQ+HU ZRUNV KDYH EHHQSXEOLVKHG LQ
VHYHUDOLQWHUQDWLRQDOMRXUQDOVDQGFRQIHUHQFHV