Several questions of Random Forest Classifier in SNAP

Hello all, I’m studying with Satellite Image Classification, especially Random Forest.
There are some questions when I perform Random Forest Classifier in SNAP.

  1. I wonder Random Forest Classifier’s mechanism in SNAP. I searched but I couldn’t get clear answers.

  2. Default value of “Number of training samples” set 5000, what does it mean?

  3. When RF process is done, we can get confusion matrix written Text file named newClassifier.
    ‘Predictive Values’ depend on pixel feature in Training vectors I guess. Then what is the criteria of ‘Actual Values’ ? How can the classifier decide TP, TN, FP, FN and accuracy ?

Thanks for reading.

Have you seen These?

An example on accuracy assessment is given in this tutorial: Landcover classification with Sentinel-1 GRD

I’ll read those. Thanks!