Week 14: Measuring Whether a Model Works

A model score is not a verdict. It is the result of comparing predictions with known outcomes under a specific evaluation procedure.

This week starts underneath the familiar word accuracy. You will build the four confusion-matrix counts, derive accuracy, precision, and recall from those counts, and then watch a decision threshold trade one type of error for another.

The central question is not “which metric is best?” It is which error matters for the decision being made?

The standalone Model Evaluation Lab lets you move a probability threshold and watch the confusion matrix change live. Your job is to defend a threshold using error consequences, not to hunt for the prettiest percentage.