Module C01 · concept player

The threshold is a cost ratio, not a knob

A classifier outputs a score, not a decision. The threshold that turns one into the other comes from what a false positive costs against what a false negative costs — slide either one and watch the cheapest threshold move off 0.5.

Cost of a false positive: 1
Cost of a false negative: 1
Threshold you are testing: 0.50
0.50threshold that minimizes cost
0.10expected cost, per case, at your threshold
+0%worse than the optimal threshold

Score distributions and expected cost, against threshold

Illustrative Gaussian score distributions, not a measured dataset. They exist so a real minimum can move on screen; no accuracy number here should be quoted as a benchmark. The principle is the citable part — C01 section 2: define the decision before the prediction, and set the threshold from the relative cost of the two error types, not from a metric's default.