Fomula yowongolera zinthu

Fomula Yosinthira Zinthu

Kubwerezabwereza kwa logistic ndi njira imodzi yotchuka kwambiri mu ziwerengero ndi sayansi ya deta yowonetsera ubale pakati pa zinthu zingapo zodziyimira pawokha (zoneneratu) ndi kusintha kodalira gulu, makamaka kwa binary (monga, inde/ayi, kupambana/kulephera, kudwala/thanzi). Mosiyana ndi kubwerezabwereza kwa mzere, komwe kumapanga ma values ​​osalekeza, kubwerezabwereza kwa logistic kumapangidwa kuti kuyerekeze kuthekera kwa chochitika, kotero zotsatira zomaliza zili pakati pa 0 ndi 1. M'nkhaniyi, tikambirana za njira yobwerezabwereza ya logistic, tanthauzo la gawo lililonse, ndi momwe tingalitanthauzire.

Nchifukwa chiyani Kubwezeretsa Zinthu Kukufunika?

Ngati tigwiritsa ntchito linear regression kuti tilosere za kuthekera, chitsanzocho chingapange ma values ​​​​omwe ali pansi pa 0 kapena kupitirira 1, zomwe sizomveka bwino pa kuthekera. Logistic regression imathetsa vutoli pogwiritsa ntchito ntchito yopanda mzere yomwe imayika zotsatira zowerengedwa (zomwe zingakhale mtengo uliwonse) ku mtengo wa kuthekera pakati pa 0 ndi 1. Ntchito yomwe imagwiritsidwa ntchito kwambiri ndi logistic kapena sigmoid function.

Mwachitsanzo, tiyerekeze kuti tikufuna kuneneratu ngati kasitomala adzasintha malinga ndi zaka zake, nthawi yomwe adalembetsa, komanso kuchuluka kwa momwe amagwiritsidwira ntchito. Zotsatira zomwe zanenedweratu zili ndi njira ziwiri zokha: kusintha (1) kapena kusasintha (0). Kusintha kwa logistic ndikoyenera kwambiri pamtunduwu.

Fomula Yoyambira ya Kubwezeretsa Zinthu

Chofunika kwambiri cha logistic regression ndikutengera mwayi \( p \) kuti \( Y = 1 \) (chochitikacho chimachitika), poganizira mtengo wa predictor variable \( X \).

Ma model a logistic regression nthawi zambiri amalembedwa m'njira ziwiri zofunika:

1) Fomu Yotheka (Sigmoid)

\[
p = P(Y=1 \mid X) = \frac{1}{1 + e^{-z}}
\]

ndi

\[
z = \beta_0 + \beta_1 X_1 + \beta_2 X_2 + \cdots + \beta_k X_k
\]

Zambiri:
– \( p \) ndi mwayi wa chochitikachi (monga: churn = 1).
– \( e \) ndi nambala ya Euler (pafupifupi 2,71828).
– \( z \) ndi kuphatikiza kolunjika kwa zinthu zoneneratu.
– \( \beta_0 \) ndiye njira yolumikizira (yokhazikika).
– \( \beta_1, \beta_2, \ldots, \beta_k \) ndi ma coefficients owongolera.
– \( X_1, X_2, \ldots, X_k \) ndi ma variable odziyimira pawokha.

Ntchito ya sigmoid imatsimikizira kuti mtengo uliwonse wa \( z \), mtengo wa \( p \) umakhalabe pakati pa 0 ndi 1.

2) Fomu Yolowera (Mavuto a Log)

Fomu ina yofunika kwambiri ndi fomu ya logit, yomwe ndi logarithm ya zovuta:

\[
\text{logit}(p) = \ln\left(\frac{p}{1-p}\right) = \beta_0 + \beta_1 X_1 + \beta_2 X_2 + \cdots + \beta_k X_k
\]

Zambiri:
– \( \frac{p}{1-p} \) amatchedwa mwayi wofanana (mwayi woyerekeza).
– \( \ln \) ndi logarithm yachilengedwe.

Fomu ya logit imafotokoza kuti logistic regression imayimira ma odds a log ngati ntchito yolunjika ya zoneneratu. Izi zimapangitsa kutanthauzira ma coefficients kukhala kosavuta, makamaka pankhani ya ma odds ratios.

Kumvetsetsa Kuchuluka kwa Odds ndi Odds

Kuti timvetse bwino njira yosinthira zinthu, tiyenera kusiyanitsa pakati pa kuthekera ndi zovuta.

– Kuthekera \( p \): mwayi woti chochitika chichitike (0 mpaka 1).
- Mwayi: kufananiza mwayi wa chinthu chomwe chingachitike kuti chisachitike:

\[
\text{odds} = \frac{p}{1-p}
\]

Chitsanzo: ngati \( p = 0{,}8 \), ndiye:

\[
\text{odds} = \frac{0{,}8}{0{,}2} = 4
\]

Izi zikutanthauza kuti chochitikachi chili ndi mwayi wochuluka nthawi zinayi kuposa zosatheka.

Mu logistic regression, coefficient \( \beta \) nthawi zambiri imatanthauziridwa kudzera mu odds ratio:

\[
\text{OR} = e^{\beta}
\]

– Ngati \( \beta > 0 \), ndiye \( e^{\beta} > 1 \): cholosera chimawonjezera mwayi wa chochitikacho.
– Ngati \( \beta < 0 \), ndiye \( e^{\beta} < 1 \): cholosera chimachepetsa mwayi wa chochitikacho. - Ngati \( \beta = 0 \), ndiye \( e^{\beta} = 1 \): palibe chomwe chimachitika pa mwayi. Mwachitsanzo, ngati \( \beta_1 = 0{,}7 \), ndiye: \[ e^{0{,}7} \pafupifupi 2{,}01 \] Izi zikutanthauza kuti kuwonjezeka kwa gawo limodzi mu \( X_1 \) kudzachulukitsa mwayi wa chochitikacho ndi nthawi pafupifupi 2,01 (poganiza kuti zosintha zina zimakhalabe zosasintha). Chitsanzo cha Chitsanzo Chosavuta cha Logistic Regression Tiyerekeze kuti tili ndi cholosera chimodzi chokha \( X \), mwachitsanzo chiwerengero cha maola ophunzirira pa sabata, kuti tinene kuti tipambana mayeso (pass = 1, fail = 0). Chitsanzo:

\[ \text{logit}(p) = \beta_0 + \beta_1 X \] Ngati zotsatira zoyerekeza ndi izi: - \( \beta_0 = -4 \) - \( \beta_1 = 0{,}8 \) Kenako: \[ z = -4 + 0{,}8X \] \[ p = \frac{1}{1 + e^{-(-4 + 0{,}8X)}} = \frac{1}{1 + e^{4 - 0{,}8X}} \] Ngati \( X = 6 \) maola ophunzirira: \[ z = -4 + 0{,}8(6) = 0{,}8 \] \[ p = \frac{1}{1 + e^{-0{,}8}} \pafupifupi 0{,}69 \] Kutanthauzira: ndi maola 6 ophunzirira pa sabata, mwayi wodutsa ndi chigoli cha pafupifupi 69%. Kuyerekeza kwa Ma Coefficient: Chifukwa Chiyani Si Njira Yogwiritsira Ntchito Ma Squares Osachepera? Mu linear regression, ma coefficient nthawi zambiri amawerengedwa pogwiritsa ntchito njira ya least squares. Komabe, mu logistic regression, ubale pakati pa zoneneratu ndi kuthekera si wofanana, kotero njira ya least squares si yabwino. Logistic regression nthawi zambiri imagwiritsa ntchito Maximum Likelihood Estimation (MLE) kuti ipeze coefficient value \( \beta \) yomwe imawonjezera mwayi wa deta yomwe yawonedwa. Mwachidule, kuthekera kwa kuwona kwa binary \( y_i \in \{0,1\} \) ndi maulosi \( p_i \) ndi: \[ L(\beta) = \prod_{i=1}^{n} p_i^{y_i}(1-p_i)^{(1-y_i)} \] Kenako nthawi zambiri imasinthidwa kukhala log-likelihood kuti ikhale yosavuta kuwerengera: \[ \ell(\beta) = \sum_{i=1}^{n} \left[ y_i \ln(p_i) + (1-y_i)\ln(1-p_i) \right] \] Mtengo wa \( \beta \) umasankhidwa kuti uwonjezere \( \ell(\beta) \). Njira zowerengera monga Newton-Raphson kapena gradient descent nthawi zambiri zimagwiritsidwa ntchito ndi mapulogalamu owerengera. Ubwino ndi Zofooka za Logistic Regression Ubwino 1. Zotsatira zake zimakhala ngati mwayi kotero ndizosavuta kuzimasulira kukhala zisankho. 2. Kutanthauzira kwa ma coefficients kumamveka bwino kudzera mu chiŵerengero cha ma odds. 3. Koyenera mavuto a magulu a binary ndipo kungapitirire ku multinomial/ordinal. Zoletsa 1. Kuganiza kuti pali ubale wolunjika pakati pa zoneneratu ndi ma odds a log, osati mwachindunji ku kuthekera. 2. Kungakhale kovuta ngati pali multicollinearity kapena deta yosalinganika kwambiri. 3. Pa machitidwe ovuta kwambiri a ubale, njira zina zosakhala zolunjika (monga, random forest kapena neural network) zingakhale zabwino kwambiri. Mapeto Fomula yosinthira ya logistic kwenikweni imaphatikiza kuphatikiza kwa mizere ya zosintha zodziwikiratu ndi ntchito ya sigmoid kuti ipange mwayi. Fomu yodziwika kwambiri ndi iyi: \[ p = \frac{1}{1 + e^{-(\beta_0 + \beta_1 X_1 + \cdots + \beta_k X_k)}} \] kapena mu mawonekedwe a logit: \[ \ln\left(\frac{p}{1-p}\right) = \beta_0 + \beta_1 X_1 + \cdots + \beta_k X_k \] Pomvetsetsa mitundu iwiri iyi ya fomula, titha kupanga zitsanzo zodziwikiratu zamavuto osiyanasiyana amitundu iwiri pamene tikutanthauzira mphamvu ya zosintha kudzera mu chiŵerengero cha odds \( e^{\beta} \). Kubwezeretsa kwa logistic kumakhalabe maziko ofunikira pakusanthula deta chifukwa ndi kosavuta, kwamphamvu, komanso kotanthauzira—ndipo nthawi zambiri ndi gawo loyamba tisanayesere zitsanzo zovuta kwambiri. Ngati mukufuna, nditha kuwonjezera kuwerengera kwachitsanzo ndi deta yaying'ono (tebulo), kapena kukhazikitsa chitsanzo cha logistic regression mu Python/R pamodzi ndi kutanthauzira kwa zotsatira.