Using the random decision forests to explain the the predictions of the black-box in decentralized setting, and at the same time we used this explanations to detect the attacks on the training process.