1
$\begingroup$

Does R-Squared for a random forest increase as the number of predictors increases? I have read over and over again that $R^{2}$ increases as you increase the number of predictors when the model is linear regression / ARIMA. Is this the case when using a random forest as a regression model for time series data?

I ask this because I am computing the out-of-sample $R^{2}$ score for a random forest trained on historical data and wondering if I should change to a different metric like MAPE.

$\endgroup$
1
$\begingroup$

Percent variance explained by a RF model does not automatically increase with increasing the number of predictors, in fact with a multitude of poor predictors it can even decline (as more 'bad' individual small trees are generated). Also note that RF has the independence assumption, which is violated in time-series data.

$\endgroup$

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.