CODEWITHSUNDEEP
rmachine-learningglmpredict
NoSe
NoSe

Reputation: 109

Predict a factor of 3 levels and return the % of each factor

I am trying to predict a three-level factor with a set of variables. The group is A, B and C.

m<-glm(as.factor(Group)~Sex+BloodType+Pressure,data=Hel,family = "binomial")

newdata <- data.frame(Sex="M",BloodType="A+", Pressure=80)

predict(m,newdata)

and this returns:

        1 
0.7133324

I would like that to give me back:

A        B        C
20.00   40.00    40.000

How can I do this? Thanks.

Upvotes: 2

Views: 1151

Answers (2)

eipi10
eipi10

Reputation: 93851

There are two issues to address:

First, glm can only do logistic regression with a two-level outcome. If you provide an outcome variable with more than two levels, it lumps all levels except the first (reference) level into the second level. To do multinomial regression, you'll need to use a different function (there are several packages that can do multinomial regression, for example, polr from the MASS package, multinom from the nnet package (see @Stupidwolf's answer), vglm from the VGAM package; also see the mlogit and gmnl packages).

Second, predict(m1, newdata) provides predictions on the log-odds scale. To get predictions on the probability scale, you would need to do predict(m1, newdata, type="response"). Similar options (though possibly with different names for the arguments and/or argument options) will need to be set to get predictions on the probability scale for multinomial models.

Upvotes: 2

StupidWolf
StupidWolf

Reputation: 46948

I guess you have to use something that allows multinomial regression like nnet?

library(nnet)
trn = sample(1:nrow(iris),100)
fit_nnet <- multinom(Species ~ ., iris[trn,])

head(predict(fit_nnet,iris[-trn,],type="prob"))
      setosa   versicolor    virginica
7  1.0000000 2.348893e-11 2.166176e-58
9  0.9999323 6.765017e-05 2.438304e-52
10 0.9999940 6.035319e-06 1.530614e-56
12 1.0000000 1.355834e-09 1.315066e-57

Upvotes: 0

Related Questions