Why `adaboost` using `gbm` package will not produce exact binary prediction values such as `{1,0}`?

Asked Sep 25 '22 at 15:43

Active Sep 25 '22 at 15:43

Viewed 147 times

My input data contains a response variable is_disease which is obviously a Factor(Yes,No). In order to use gbm package to perform a adaboost tree, I convert the {"yes","no"} to {1,0}. After I build a model

model1 <- gbm(train, is_disease ~ .,distribution = "adaboost",n.trees = 50),

then I make a prediction

p1 <- predict.gbm(m1, test, n.trees = 50,type = "response"). I found that the predicted values is between [0,1],but most of the predicted values look like 0.04699193 0.05956444. However, why it is not producing any predictions like exactly 0 or 1? Since as a classification problem, I need more binary values rather than floats between [0,1].

asked Sep 25 '22 at 15:43

nobodyishere

2

The decimal numbers can presumably be interpreted as probabilities of is_disease. These are much better than hard classifications. You can always compare them to some threshold to turn them into hard classifications, but I would recommend you think about your problem before you do so. – Stephan Kolassa Sep 25 '22 at 15:47
1

It does exactly what its documentation says. – Tim Sep 25 '22 at 15:48

Why `adaboost` using `gbm` package will not produce exact binary prediction values such as `{1,0}`?

0 Answers0