- I-VIF ithola i-multicollinearity kumamodeli okubuyela emuva ukuze igweme ukuphambuka kwama-coefficients.
- I-GVIF isetshenziswa lapho imodeli ihlanganisa iziguquguquko zezigaba ezinamazinga amaningi.
- Ukusetshenziswa kobuchwepheshe kungasekelwa emitatsheni yezincwadi ye-C# noma ngokuhlanganiswa kwamaphakheji e-R akhethekile.
Uma ungena ekuhlaziyweni kwedatha kanye nezibalo , cishe usuthole umqondo we-multicollinearity. Ngokuyisisekelo, lokhu kwenzeka lapho iziguquguquko ezimbili noma ngaphezulu ezizimele kumodeli yokubuyela emuva zihlobene kakhulu kangangokuthi uhlelo luyadideka futhi lungakwazi ukunquma ukuthi iyiphi ebangela umphumela, ekugcineni kuphambukise izibikezelo zakho.
Ukuze kulungiswe lokhu kudideka, i-Variance Inflation Factor, eyaziwa ngothando ngokuthi i-VIF, iyasetshenziswa. Le metric isivumela ukuthi sinqume ukuthi i-variable ayisebenzi kahle yini nokuthi ikhulisa yini i-variance yama-coefficients alinganisiwe, okusho ukuthi isitshela ukuthi sinedatha ehambisanayo yini.
Iyini ngempela i-VIF futhi isebenza kanjani?
I-VIF iyithuluzi lokuxilonga elilinganisa ukuthi ukuhlukahluka kwe-regression coefficient kukhula kangakanani ngenxa yokuxhumana nezinye izinto eziguquguqukayo kumodeli. Ngobuchwepheshe, kubalwa ngokwenza i-linear regression lapho enye yezinto ezizimele isebenza njenge-dependent variable ngokuphathelene nezinye. Uma umphumela ungu-1, akukho ukuhlobana; uma kungaphezu kuka-10, sivame ukuba nenkinga enkulu ye-collinearity okufanele ixazululwe.
Uma sisebenza ngeziguquguquko zezigaba ezinamazinga angaphezu kwamabili, i-VIF ejwayelekile iyasilela, futhi kumelwe sishintshele ku-GVIF, noma i-Generalized Variance Inflation Factor . Lokhu kulungiswa kubalulekile ukuze ukubala kuhambisane, kusetshenziswa ifomula yokumisa ejwayelekile evame ukuba yi-(G)VIF ephakanyiswe emandleni ka-1 ehlukaniswe kabili ngamadigri enkululeko.
Ukuqaliswa kanye namathuluzi athuthukile
Nakuba izibalo zezibalo zisebenza emhlabeni wonke, amathuluzi okuhlaziya idatha akhona azenzakalela le nqubo. Isibonelo, ezindaweni ezifana ne-AlteryxOne (izinguqulo 2025.1 nezakamuva), ithuluzi elithile le-VIF liyatholakala ku-Community Gallery. Le nsizakalo ingakhiqiza imibiko efingqiwe ye-coefficient eningiliziwe yanoma yikuphi ukuhlukahluka, ngaphandle kwe-intercept, okuhlala kugcinwa inani elingu-1.
- Ukusekelwa kwemodeli: Ingasetshenziswa ku-linear, logistic, counting, kanye ne-gamma regressions.
- Ukuthembela ku-R: Eziningi zalezi zindlela zokusebenzisa zisebenzisa izindlela ze-R zomthombo ovulekile, ikakhulukazi iphakheji
vif, ukucubungula idatha. - Imikhawulo yobuchwepheshe: Kubalulekile ukuqaphela ukuthi izindlela ezithile, njenge-Revo ScaleR, azigcini ulwazi oludingekayo ukuze kubalwe lezi zici, ngakho-ke azihambisani nalezi macros.
Ukubala i-VIF ku-C# kanye nezilimi zokuhlela
Ukuze sisebenzise lokhu ku-C#, akukho msebenzi wendabuko olimini oluyisisekelo, ngakho-ke kumelwe sithembele emitatsheni ye-algebra eqondile njenge-Math.NET Numerics. Inqubo ihilela ukwakha i-matrix yokuxhumana nokubala okuphambene nayo, noma ukusebenzisa ama-regressions asekelayo ku-variable ngayinye ezimele.
Ukugeleza okunengqondo ku-C# kuzohlanganisa ukuhlukanisa i-predictor variable ngayinye, ukuyiphatha njengenhloso yokubuyela emuva okuqondile ngokumelene nezinye izibikezeli, ukuthola i-coefficient of determination R², nokusebenzisa ifomula 1 / (1 - R²). Uma umhleli wezinhlelo efuna i- implementation eqinile , ikhambi elifanele ukuhlanganisa ama-wrappers abiza i-Python yokuhlaziywa kwedatha noma ama-script e-R, njengoba ukuphathwa kwe-matrix kanye nokuqinisekiswa kwe-collinearity kuvuthiwe kakhulu kulezo zilimi.
Ukuba nokulawula okuqinile phezu kwe-VIF kukuvumela ukuthi uhlanze imodeli ngokususa iziguquguquko ezingafuneki, okungagcini nje ngokuthuthukisa ukunemba, kodwa futhi kwenza imodeli ihunyushwe kalula futhi isebenze kahle ngokubala, ivimbele umsindo wezibalo ukuthi ungafihli ubudlelwano bangempela phakathi kwedatha.
