Bridging the Gap: Unifying the Training and Evaluation of Neural Network Binary Classifiers