Achieve 0.95 precision via thresholding

Suppose you are shipping a probabilistic binary classifier for a heavily imbalanced dataset containing 80,000 records, with positive examples accounting for roughly 6%. The release requirement is strict: the model's positive-class precision on a held-out test set must be at least 0.95. Complete the following steps: Fit any model that emits probabilities. Using a separate validation split, scan the precision-recall curve and choose the smallest decision threshold $$\tau$$…