forum.alglib.net

hello,
i use decision-forest for a 2 class classification and when building a forest it says that miss-classification is about .02 in final report. but when i try to process each training data with process function the probabilities return to me is highly biased to class 0 (avg .95) and class 1 is rare (avg .05). is it possible i train the forest incorrectly or use process function incorrect?
is there any samples to see how to make and use decision-forest from a double array of size 10000 instances?
i train data with a real_2d array with columns nfeature+1 and test data with real_1d array with columns nfeature.
thanks in advance

Is it possible that class 1 is really rare in your training set and can not be easily separated from class 0? When two classes can not be separated, decision forest returns expected class membership probabilities. Say, if one sample of class 1 is closely surrounded by 20 samples of class 0, then forest will return probability equal to ~0.05.