0
$\begingroup$

Random forests is an ensemble of trees that learns the hidden patterns in the data. I have mostly tried doing some feature-engineering before running the Random Forest model but is it required or the model takes care of this?

$\endgroup$
4
$\begingroup$

Absolutely it is. Regardless of the machine learning algorithm you are using, the quality of your features is paramount. In many instances, feature engineering (say, using domain knowledge to restructure/recode variables) creates information that was simply not there previously. To the extent that this effort introduces useful class separation, your model will very likely perform better. Random forests, or any other algorithm that I am aware of, cannot create this information without your intervention.

$\endgroup$
  • $\begingroup$ I also think that it is but perhaps it is better to qualify this statement. Specifically, I think that feature engineering in terms of interaction terms is not that important for RFs (although it can still help) since RFs capture these to a certain extent by themselves. Do you agree? $\endgroup$ – Poete Maudit Jun 13 at 15:41
0
$\begingroup$

I would say yes.RandomForest uses the given features not create new features of its own from the given set of features like Neural Network. If you add relevant features or drop insignificant features it will improve the model performance.

$\endgroup$
0
$\begingroup$

Absolutely yes, but the damage caused by splitting on noise features will vary depending on other RF parameters (nodesize, num. of variables selected per split, etc).

If your good features dominate, the damage will be small, but removing useless variables will always improve performance (both accuracy and processing time).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.