See the phrase "random forests is to improve performance by reducing the model variance", had been with random forests for classification of CART, a random forest for a test sample produce a classification results, classification of only two possible results: classification and correct classification error; That for a set of test sample set, can be calculated is the random forest for this group of sample classification accuracy (error), then the "variance" how exactly calculated? Where I can see it from the variance is reduced or increased?