The output is generated by using multiple decision trees. The trees are trained by splitting the training data randomly. So the output of each desicion tree is compared and the overall output is generated