a reasonable answer is given here: Maximum Likelihood Classifier results are different each time when re-run on the same training data
Edit: I didn’t see that you are using a random forest classifier. @MCG is right - every result will be slightly different. But they become more similar with an increasing number of trees.