Important considerations are not included in any of these discussions. The procedures discussed above invite inappropriate thresholding and utilize improper accuracy scoring rules (proportions) that are optimized by choosing the wrong features and giving them the wrong weights.
Dichotomization of continuous predictions flies in the face of optimal decision theory. ROC curves provide no actionable insights. They have become obligatory without researchers examining the benefits. They have a very large ink:information ratio.
Optimum decisions don't consider "positives" and "negatives" but rather the estimated probability of the outcome. The utility/cost/loss function, which plays no role in ROC construction hence the uselessness of ROCs, is used to translate the risk estimate to the optimal (e.g., lowest expected loss) decision.
The goal of a statistical model is often to make a prediction, and the analyst should often stop there because the analyst may not know the loss function. Key components of the prediction to validate unbiasedly (e.g., using the bootstrap) are the predictive discrimination (one semi-good way to measure this is the concordance probability which happens to equal the area under the ROC but can be more easily understood if you don't draw the ROC) and the calibration curve. Calibration validation is really, really necessary if you are using predictions on an absolute scale.
See the Information Loss chapter in Biostatistics for Biomedical Research and other chapters for more information.