Is there any way to set a threshold value that can be set for voting results for Random Forest Classifier in sklearn package in order to prevent any misclassification for incoming test data.
$\begingroup$
$\endgroup$
1
-
1$\begingroup$ Your question is too broad. In any case, I recommend you to check this link $\endgroup$– RipsteinApr 9, 2018 at 14:20
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Yes and the most important parameter is the tree depth. It's a pre-pruning technique that allows to prevent overfitting. Specifically for sklearn is:
estimator.tree_.max_depth
I suggest you to perform GridSearch on max_depth:
params = {'max_depth':[1,50]}
gs = GridSearchCV(DecisionTreeClassifier(), params)
gs.fit(X,y)
where X is you training set containing instances and y are the labels.
There are packages that also support post-pruning like this one https://svaante.github.io/decision-tree-id3/
