Skip to main content
3 votes

If R2 is not appropriate for non-linear ML algorithms such as Random Forests, can a Pearson or Spearman correlation be used as performance metric?

First, note that $R^2$ is not an invalid measure of performance for nonlinear models. Let's look at the equation. $$ R^2=1-\left(\dfrac{ \overset{N}{\underset{i=1}{\sum}}\left( y_i-\hat y_i \right)^2 }...
Dave's user avatar
  • 64.4k
1 vote

If R2 is not appropriate for non-linear ML algorithms such as Random Forests, can a Pearson or Spearman correlation be used as performance metric?

I would argue that ordinal correspondence between the predicted and observed values is indeed a reasonable way to express the performance of a non-linear model. It is in fact common to see Somers' D ...
Frans Rodenburg's user avatar
1 vote

Evaluating a classifier's performance on different groups of subjects

Since precision and recall are basically binary outcomes, you could compare the precisions and recalls directly. However, with the given sample sizes I think you're probably going to have a poor time ...
Demetri Pananos's user avatar

Only top scored, non community-wiki answers of a minimum length are eligible