The default probability cutoff for predictive modeling algorithms is p = .50. That is, if the probability calculated for an instance is greater than or equal to 0.50, the predicted value is 1 or Yes. The default cutoff minimnizes overall error, which is the combination of false positives and false negatives. However, the cost of false positive and false negative are usually not equal. In this situation, you may want to avoid one type of error more than the other. In this case, you may want to change the probability cutoff to a different value.

When the probability cutoff changes, so does the confusion matrix. Because the ROC chart is derived from trying all possible probability cutoffs, the ROC curve and area under the curve does not change when the cutoff value changes.