Isilinganiso Sokunquma: Incazelo, Ukubala, kanye Nokusetshenziswa
I-Pendahuluan
I-coefficient of determination, evame ukufanekiswa ngokuthi \(R^2 \), ingumqondo wezibalo odlala indima ebalulekile ekuhlaziyweni kwedatha. Inikeza isilinganiso sokuthi idatha ebonwe ingachazwa kahle kangakanani ngemodeli esetshenzisiwe. Nakuba isetshenziswa kabanzi ekubuyiseleni okuqondile, i-coefficient of determination iphinde ibe nezinhlelo zokusebenza kwezinye izimo ezahlukene lapho kusetshenziswa khona ukubikezela kwezibalo kanye nokulingisa.
Lesi sihloko sihlose ukuchaza incazelo ye-coefficient of determination, indlela yayo yokubala, kanye nokunikeza izibonelo zezinhlelo zokusebenza zayo zangempela. Ukuqonda lo mqondo kuzothuthukisa ikhono lethu lokwenza ukuhlaziywa kwezibalo okunenjongo kakhudlwana.
Ukuqonda i-Coefficient of Determination
I-coefficient of determination (\( R^2 \)) iyinani eliphakathi kuka-0 no-1, elibonisa isilinganiso sokuguquguquka ku-dependent variable esingachazwa yi-variables ezimele kumodeli yokubuyela emuva. Inani lika-\( R^2 \) eliseduze no-1 libonisa ukuthi i-variables ezimele ezikhethiwe ziyakwazi ukuchaza iningi lokuguquguquka ku-dependent variable, kuyilapho inani lika-\( R^2 \) eliseduze no-0 libonisa ukuthi imodeli ayanele ngokwanele ekuchazeni ukuguquguquka kwedatha.
Ngokwezibalo, i-coefficient of determination ingabonakaliswa ngefomula:
\[ R^2 = 1 – \frac{SSR}{SST} \]
Di mana:
– \( SSR \) yi-Total Sum of Squares Residual (isamba sezikwele zezinsalela noma amaphutha okubikezela)
– \( SST \) yi-Total Sum of Squares (isamba sezikwele ze-variable exhomeke kuyo)
Indlela Yokubala I-Coefficient Yokunquma
Ukuze siqonde kangcono ukuthi i-coefficient of determination ibalwa kanjani, ake sicacise izinyathelo.
Izinyathelo Zokubala
1. Bala inani elibikezelwe (\( \hat{y} \)):
Leli nani elibikezelwe litholakala kumodeli yokubuyela emuva esiyidalile. Isibonelo, uma imodeli yokubuyela emuva ilula, khona-ke isimo sayo siyilokhu:
\[ \hat{y} = \beta_0 + \beta_1 x \]
2. Bala insalela (\( e \)):
Okusele umehluko phakathi kwenani elibonwe (\( y \)) kanye nenani elibikezelwe (\( \hat{y} \)):
\[ e = y – \hat{y} \]
3. Bala i-SSR (Isamba Sezinsalela Eziyisikwele):
I-SSR iyisamba sezikwele zezinsalela:
\[ SSR = \sum (y – \hat{y})^2 \]
4. Bala i-SST (Isamba Sezikwele):
I-SST iyisamba sezikwele zomehluko phakathi kwamanani abonwe (\( y \)) kanye nesilinganiso salawo manani abonwe (\( \bar{y} \)):
\[ SST = \ isamba (y – \ ibha{y})^2 \]
5. Bala i-Coefficient of Determination (\( R^2 \)):
I-coefficient of determination ibalwa kusetshenziswa ifomula eshiwo ngenhla:
\[ R^2 = 1 – \frac{SSR}{SST} \]
Ngalezi zinyathelo, singakhiqiza inani elingu-\( R^2 \) elichaza ukuthi imodeli yethu ichaza kahle kangakanani ukuguquguquka kwedatha.
Ukuqonda Inani le-\( R^2 \)
Inani le-\( R^2 \) lingahluka kakhulu kuye ngomongo kanye nobunzima bemodeli esetshenzisiwe. Nazi ezinye iziqondiso zokuhumusha inani le-\( R^2 \):
– \( R^2 \cishe 0 \):
Lokhu kubonisa ukuthi imodeli yokubuyela emuva ayikwazi ukuchaza ukuguquguquka kwedatha. Iziguquguquko ezizimele ezisetshenzisiwe zingase zingabalulekanga kuguquguqukayo encike kuyo, noma kungaba ngenxa yokuguquguquka okukhulu kwedatha yochungechunge lwesikhathi.
– \( 0 < R^2 < 0.3 \): Imodeli inekhwalithi ephansi kakhulu yokuchaza, kodwa kusekhona ulwazi oluthile olungatholakala mayelana nobudlelwano phakathi kweziguquguquko.
- \( 0.3 \leq R^2 < 0.7 \): Leli nani libonisa ukuthi imodeli inekhwalithi yokuchaza ephakathi nendawo. Imodeli iwusizo impela kodwa kusenendawo yokuthuthukisa. - \( 0.7 \leq R^2 < 1 \): Imodeli inekhwalithi yokuchaza ephezulu. Iningi lokuguquguquka kwe-variable exhomeke kuyo lichazwa yiziguquguquko ezizimele. - \( R^2 \approx 1 \): Lokhu kubonisa ukuthi imodeli inekhwalithi yokuchaza ephezulu kakhulu. Kodwa-ke, lokhu kufanele futhi kucatshangelwe ngoba kungabonisa ukulingana ngokweqile, lapho imodeli iyinkimbinkimbi kakhulu futhi ingasakwazi ukujwayelana. Izicelo Zomhlaba Wangempela I-coefficient yokunquma isetshenziswa emikhakheni ehlukahlukene, kusukela kwisayensi yezenhlalo kuya kwisayensi yemvelo. Nazi ezinye izibonelo eziqondile zokusetshenziswa kwe-\( R^2 \): 1. Ezomnotho: Ekuhlaziyweni kwezomnotho, i-coefficient yokunquma isetshenziswa ukuhlola ukuthi imodeli yezomnotho ikwazi kahle kangakanani ukuchaza ubudlelwano phakathi kweziguquguquko ezifana nemali engenayo, ukusetshenziswa, kanye nokutshalwa kwezimali. 2. Izibalo ze-Bio: Ocwaningweni lwezokwelapha, i-\(R^2 \) isetshenziselwa ukuhlola ukusebenza kahle kobudlelwano phakathi komthamo wemithi kanye nempendulo yesiguli. Imodeli enhle izoba ne-\(R^2 \ ephezulu), okubonisa ukuthi umthamo wemithi ungachaza iningi lokuguquguquka kwempendulo yesiguli. 3. Isayensi Yezemvelo: Ekuboniseni isimo sezulu, i-\(R^2 \) ingasetshenziswa ukuhlola ubudlelwano phakathi kwezici zesimo sezulu ezifana nemvula, izinga lokushisa, kanye nomswakama. I-coefficient ephezulu yokunquma ikhombisa ukuthi imodeli yesimo sezulu esetshenzisiwe inhle kakhulu ekuchazeni ukushintshashintsha kwesimo sezulu. 4. Ibhizinisi Nokumaketha: Ekuhlaziyweni kokumaketha, i-\(R^2 \) ingasetshenziswa ukuhlola ukuthi imodeli yokubuyela emuva ichaza kahle kangakanani ubudlelwano phakathi kwezindleko zokukhangisa nokuthengisa. Inani eliphezulu le-\(R^2 \) libonisa ukuthi izindleko zokukhangisa ziyisibonakaliso esihle sokuthengisa. Imikhawulo ye-Coefficient of Determination Nakuba i-coefficient of determination iyithuluzi eliwusizo kakhulu, futhi inemikhawulo eminingana okudingeka icatshangelwe: 1. Ayilinganisi Imbangela: Inani eliphezulu le-\(R^2 \) alibonisi ukuthi i-variable ezimele ibangela ukuthi i-variable exhomeke kuyo ishintshe. Libonisa kuphela ubudlelwano obuqondile phakathi kwalokhu okubili. 2. Ingathinteka yi-Overfitting: Imodeli eyinkimbinkimbi kakhulu ingaba ne-\(R^2 \) ephezulu kakhulu kodwa ingabi yinto evamile kweminye idatha. Ngakho-ke, kubalulekile ukucabangela nezindlela zokuqinisekisa le modeli, njengokuqinisekiswa okuphambene. 3. Ayinikezi Ulwazi Ngekhwalithi Yomuntu Ngamunye Yama-Predictors: I-coefficient of determination ayinikezi ulwazi mayelana nomnikelo womuntu ngamunye we-variable ngayinye ezimele kumodeli ye-multivariate. Isiphetho I-coefficient of determination (\( R^2 \)) iyithuluzi lezibalo elibaluleke kakhulu lokuhlola ikhono lemodeli lokuchaza ukuguquguquka kwedatha. Ngokuqonda ukuthi ibalwa futhi ihunyushwa kanjani, singenza kangcono ukuhlaziywa kwedatha okunengqondo. Kubalulekile ukusebenzisa njalo i-\( R^2 \) njengenye yamathuluzi amaningi ezibalo, ukuqonda ukulinganiselwa kwayo, nokuyisebenzisa kanye nezinye izindlela zokuqinisekisa ukuthola imiphumela ephelele nenembile. Lokhu kuphetha isihloko mayelana ne-coefficient of determination. Ngethemba ukuthi, izoba usizo ekuqondeni kangcono nasekusebenziseni lo mqondo ekuhlaziyweni kwedatha yakho.