IV. DISCUSSION
Our neural network classification model, which was trained via the two-stage back-propagation approach, was used to diagnose and classify malignant breast cancer and benign disease for breast biopsy outcome predictions. The classification model was trained and tested using the pattern dataset based on the resubstitution method. The test accuracy of our classification model in distinguishing malignant breast cancer and benign disease was 89.64%. Accordingly, the sensitivity was 89.33%, specificity was 89.93%, and precision was 89.33%. The estimated area under the ROC curve was 0.9626 and its standard error was 0.0069. This is to say that if mammography was used to obtain the 5 attributes (referred to as BIRADS, age, shape, margin, and density) from a new patient, the diagnostic results via our classification model would be 89.64% accurate in diagnosing and classifying malignant breast cancer and benign disease. In addition, with the high area of the ROC curve results (0.9626±0.0069), our classification model can provide a consistently high accuracy in diagnosing malignant breast cancer and benign disease for breast biopsy outcome predictions.