Where do you put the threshold?

A classifier outputs a probability. Turning it into a decision needs a cut-off — and that cut-off is a choice, not a property of the model. Press play to sweep it, then set it yourself. Then change how rare the positive class is.

Decision threshold
predict positive when score ≥ 0.500
Positive class prevalence
Confusion matrix
Predicted positivePredicted negative
Actually positive 0True Positive 0False Negative · Type II
Actually negative 0False Positive · Type I 0True Negative
Metrics at this threshold
Precision
–
Recall
–
Specificity
–
False positive rate
–
F1 score
–
Accuracy
–
ROC curve — AUC –
Precision–recall curve — AUPRC –
A model that never predicts the positive class at all scores – accuracy here.

The scores the model gives each class never change — only how many negatives there are. The ROC curve and its AUC barely notice. Precision, and the area under the precision–recall curve, collapse.