Izindlela Zokufaka Izibalo Kuzibalo
Ekwenzeni izibalo kanye nokuhlaziywa kwedatha, inkinga yokuntuleka kwedatha cishe njalo iphakama. Idatha ingalahleka ngenxa yokuthi abaphenduli abayiphenduli imibuzo ethile, amaphutha okurekhoda, ukuphazamiseka kwenzwa, ukonakala kwedatha ngesikhathi sokukhipha, noma ngenxa yenqubo yokuhlanganisa imithombo eminingi yedatha engahambisani ngokugcwele. Uma ingaphathwanga kahle, idatha elahlekile ingonakalisa ikhwalithi yokuhlaziywa , inciphise amandla okuhlolwa, futhi iholele ngisho naseziphethweni ezichemile. Enye yezindlela ezivame kakhulu zokusingatha idatha elahlekile ukufakwa kwemininingwane, okuhlanganisa ukugcwalisa amanani alahlekile ngamanani alinganisiwe ngokusekelwe kolwazi olutholakalayo.
Kungani Kubalulekile Ukufaka I-Imputation?
Kunezizathu eziningana zokuthi kungani ukufakwa kwedatha kuvame ukukhethwa kunokususa idatha engekho. Okokuqala, ukususa imigqa/ukubuka okuqukethe amanani angekho (isb., ukususwa kohlu) kunganciphisa kakhulu usayizi wesampula, ikakhulukazi lapho iphesenti ledatha elahlekile libalulekile. Okwesibili, uma idatha ingatholakali ngokungahleliwe, ukususwa kungaletha ukucwasa. Okwesithathu, ama-algorithm amaningi ezibalo noma okufunda komshini adinga idatha ephelele , okwenza ukufakwa kwedatha kube yisinyathelo esilula sokucubungula.
Kodwa-ke, ukufaka imininingwane akugcini nje ngokufaka imininingwane "ekugcwaliseni izikhala." Indlela ekhethiwe kumele icabangele indlela yolwazi engekho, isakhiwo sezinto eziguquguqukayo, kanye nezinhloso zokuhlaziya. Ukungafaki imininingwane okungekuhle kungenza "ukukhohlisa" imodeli, kunciphise ukuhlukahluka, futhi kwenze imiphumela ibonakale iqinisekile kakhulu kunalokho eyikho ngempela.
Indlela Yokuchitha Idatha Elahlekile
Ezincwadini zezibalo, idatha engekho ivame ukuhlukaniswa ngezindlela ezintathu eziyinhloko:
1. I-MCAR (Engekho Ngokuphelele Ngesikhathi Esingahleliwe): amathuba okulahleka kwedatha ancike kunoma yiziphi iziguquguquko, ezibonwe noma ezingabonwanga. Isibonelo, uhlu lwemibuzo olulimele ngengozi.
2. MAR (Okungekho Ngesikhathi Esingahleliwe): amathuba okulahleka kwedatha ancike ku-variable ebonwe, kodwa akuxhomekile enanini elingekho ngokwalo ngemva kokulawula ezinye izinto eziguquguqukayo. Isibonelo, abaphenduli abasebasha banamathuba amaningi okuphuthelwa yimibuzo yemali engenayo, kodwa ubudala buyatholakala.
3. I-MNAR (Engekho Ngesikhathi Esingahleliwe): Amathuba okulahleka kwedatha ancike enanini elingekho ngokwalo. Isibonelo, abantu abanemali engenayo ephezulu kakhulu bavame ukungadaluli imali yabo engenayo.
Ngokuvamile ukufakwa kwedatha kuphephile ngaphansi kwe-MCAR/MAR. I-MNAR ivame ukudinga imodeli echaza ngokusobala idatha engekho noma ukuhlaziywa kokuzwela.
Indlela Elula Yokufaka
1. Ukufakwa Okumaphakathi/Okumaphakathi/Kwemodi
Indlela elula kakhulu ukufaka amanani angekho esikhundleni se-mean noma i-median yezinguquko zezinombolo, kanye nemodi yezinguquko zezigaba. Izinzuzo yilezi: kulula, kuyashesha, futhi kuvame ukusebenza njengesisekelo. Okubi yilezi: kunganciphisa ukuhlukahluka futhi kuphambukise ukusatshalaliswa kwedatha, ikakhulukazi uma idatha ingalingani noma iqukethe izinto ezingaphandle. I-median ivame ukuba namandla kakhulu uma iqhathaniswa ne-outliers kune-average..
2. Ukugxekwa Okuqhubekayo
Amanani angekho agcwele i-constant ethile, njengo-0, -1, noma ilebula ethi "Akwaziwa." Lokhu kuyasiza uma inani linencazelo ethile (isb., "akukho kuthengiselana"), noma lapho imodeli idinga ukunikezwa inkomba eyengeziwe ukuze igqamise ukuntuleka. Kodwa-ke, ukukhetha i-constant engahleliwe kungaletha amaphethini angafanele.
3. Ukufakwa Kwe-Hot Deck
Ku-hot deck, amanani angekho agcwaliswa kusetshenziswa amanani avela kokunye ukubonwa "okufanayo" (abanikeli) ngokusekelwe ezinguqukweni eziningana ezibalulekile. Le ndlela idumile ezinhlolovo. Ama-hot deck agcina amanani angokoqobo ngoba abamba amanani angempela edatha, kodwa imiphumela iyazwela encazelweni "yokufana" futhi ingaveza ukuhlukahluka phakathi kwesampula.
Indlela Yokufaka Esekelwe Kumodeli
4. Ukuphazamiseka Kokuhlushwa
Amanani angekho abikezelwa kusetshenziswa imodeli yokubuyela emuva kwezinye izinto eziguquguqukayo. Kuma-variable ezinombolo, ukuhlehla okuqondile kungasetshenziswa; kuma-variable ezigaba, kungasetshenziswa. ukuhlehla kwe-logistic noma i-multinomial. Inzuzo: isebenzisa ubudlelwano phakathi kweziguquguquko. Ububi: ukusebenzisa amanani abikezelwe anqunyiwe kuphela kuvame ukunciphisa ukuhlukahluka ngoba wonke amanani aqageliwe awela ngqo emgqeni obikezelwe. Ukuze kulungiswe lokhu, izingxenye ezingahleliwe (isb., izinsalela) zivame ukungezwa ukuze kube ngokoqobo okukhulu.
5. k-Omakhelwane Abaseduze (kNN) Imputation
Indlela ye-kNN igcwalisa amanani angekho ngokusekelwe esilinganisweni (noma ekuvoteni) somakhelwane abaseduze be-k. Ukusondelana kuvame ukulinganiswa ngebanga le-Euclidean noma enye i-metric ngemuva kokuthi idatha ijwayelekile. Izinzuzo zayo zifaka phakathi ukuguquguquka kanye nokucatshangwa kokungabi nobudlelwano obuqondile. Okubi kufaka phakathi ukubiza kakhulu ngamasethi amakhulu edatha, ukuzwela esikalini esiguquguqukayo, kanye nokuwohloka kokusebenza kwedatha ephezulu (isiqalekiso sobukhulu).
6. Ukulindela-Ukukhulisa (EM)
Indlela ye-EM ilinganisa amapharamitha emodeli (isb., isilinganiso kanye ne-covariance yedatha evamile ye-multivariate) ngokuphatha amanani angekho njengezinto eziguquguqukayo ezifihliwe. Isinyathelo E sibala ngokuphindaphindiwe ukulindela kwamanani angekho ngokusekelwe kumapharamitha wamanje, bese isinyathelo M sibuyekeza amapharamitha ngokusekelwe kudatha "ephelele" elindelwe. I-EM iqinile ezicatshangweni ezithile zokusabalalisa, kodwa ingaba yinkimbinkimbi futhi incike ekunembileni kwezicatshangelwe zemodeli.
Ukufakwa Kwemininingwane Eminingi: Indinganiso Yegolide Ezimweni Eziningi
7. Ukufakwa Kwemininingwane Eminingi (i-MI)
I-Multiple Imputation ibhekwa njengenye yezindlela ezisekelwe kakhulu ku-MAR. Esikhundleni sokukhiqiza isethi yedatha eyodwa ephelele, i-MI ikhiqiza amasethi edatha amaningi (isb., 5–20) anezincazelo ezahlukene ezibonisa ukungaqiniseki. Isethi ngayinye yedatha ihlaziywa ngokwehlukana, bese imiphumela ihlanganiswa kusetshenziswa imithetho kaRubin ukuthola izilinganiso ezisebenzayo kanye namaphutha ajwayelekile.
Izinzuzo ze-MI:
- Ivumela ukungaqiniseki kokusolwa.
- Kunembile kakhulu ekuqageleni kwezibalo (izikhawu zokuzethemba, ukuhlolwa kwe-hypothesis).
- Iguquguqukayo ngezinhlobo ezahlukene zeziguquguquko.
Imikhawulo yayo:
- Ukuqaliswa okuyinkimbinkimbi kakhulu.
- Kudinga ukuqagela okwanele kanye nemininingwane yemodeli yokucatshangelwa.
– Uma kukhona okuntulekayo ku-MNAR, i-MI ejwayelekile isengabandlululwa.
Ukufakwa Kochungechunge Lwesikhathi kanye Nedatha Yesikhala
Kudatha yochungechunge lwesikhathi, amanani angekho avame ukuhlobana kakhulu namanani awo angaphambilini nalandelayo. Izindlela ezifana nokuhlanganiswa okuqondile, ama-spline, izihlungi ze-Kalman, noma amamodeli e-ARIMA/State Space zivame ukusetshenziswa. Kudatha yendawo, izindlela ezifana namamodeli e-kriging kanye nendawo zingasebenzisa ukusondela kwendawo. Lezi zindlela ziyasebenza lapho isakhiwo sesikhathi/sendawo sibusa, kodwa kufanele kuqashelwe izinguquko ezingazelelwe (isb., ukushaqeka kwezomnotho) okungenza ukuhunyushwa okulula kudukise.
Imikhuba Emihle Ekukhetheni Izindlela Zokufaka I-Imputation
1. Yenza ukuhlola kwedatha engekho: hlola iphesenti lamanani angekho, iphethini yokuntuleka, nokuthi ingabe ukuntuleka kuhlobene yini nenguquko ethile.
2. Hlukanisa idatha yokuqeqesha kanye neyokuhlola: yenza ukucashunwa "ngokufunda" kudatha yokuqeqesha kuphela, bese uyisebenzisa kudatha yokuhlola ukuze ugweme ukuvuza kwedatha.
3. Cabanga ngohlobo oluguquguqukayo: olunezinombolo, olunezigaba, olunezilinganiso ezijwayelekile, noma oluxubile; indlela efanele iyahluka.
4. Sebenzisa izinkomba zokuntuleka: ngezinye izikhathi ulwazi lokuthi inani alikho ngokwalo lubikezela; ukwengeza iziguquguquko zezinkomba kungathuthukisa ukusebenza kwemodeli yokubikezela.
5. Hlola umthelela wokufakwa kwe-inputation: qhathanisa ukusatshalaliswa ngaphambi/ngemuva kokufakwa kwe-inputation, hlola ukuthi ukuhlukahluka kwehlile yini, bese uqinisekisa imodeli.
6. Beka phambili i-MI ukuze uthole umbono: lapho umgomo wokuhlaziya kuwukulinganisa amapharamitha kanye nokuhlolwa kwezibalo, ukunikezwa okuningi kuvame ukufaneleka kakhulu kunokunikezwa okukodwa.
Isiphetho
I-Imputation iyisici esibalulekile semisebenzi yesimanje yezibalo yokusingatha idatha engekho. Izindlela ezilula ezifana ne-mean/median zingaba usizo njengesisekelo noma ukuntuleka okuncane, kodwa zivame ukulimaza isakhiwo sedatha kanye nokungaqiniseki. Izindlela ezisekelwe kumodeli ezifana ne-regression, i-kNN, kanye ne-EM zisebenzisa ubudlelwano phakathi kwezinguquko, kuyilapho i-Multiple Imputation inikeza uhlaka oluqinile lokuphetha ngokubala ukungaqiniseki. Ukukhetha indlela engcono kakhulu kuncike endleleni yedatha engekho (i-MCAR/MAR/MNAR), inhloso yokuhlaziya (ukubikezela vs. ukuphetha), kanye nezici zedatha (uchungechunge lwesikhathi, indawo, exubile). Ngendlela efanele, i-imputation isiza ukugcina ubuqotho bokuhlaziya futhi iveze iziphetho ezinokwethenjelwa kakhudlwana.
