1

The scores are $R^2$ values of around 20 machine learning classifiers. Which of the scipy statistical tests methods would be appropriate here? (https://docs.scipy.org/doc/scipy/reference/stats.html#statistical-tests) I'd like to see the best scores, for example, "according to a t-test with p <.05". For example, the classifier results might look like this:

Classifier Score
RF 0.42
MLR 0.12
ANN 0.71
SVR 0.72
... ...

The best scores according to a t-test with p <.05 might be something like "0.71" and "0.72".

ds02
  • 11
  • Welcome to Cross Validated! What makes you see a t-test as appropriate for your task? I don’t see it but am open to learning. // Perhaps you can expand on what exactly you’re doing. For instance, depending on how you calculate $R^2$, you might not be shielding against overfitting the training data. (And if you’re a Python user, you’re probably even calculating $R^2$ with a function with which I disagree, though that is less of a concern of mine right now.) – Dave Oct 06 '22 at 05:16
  • r2 are probably not enough, you will need target/prediction for each data point in the test set or at least sample size. – rep_ho Oct 06 '22 at 08:42

0 Answers0