- Block 22 — Guided laboratories
- Objective
- build, evaluate and operate a card-fraud classifier on an authorization
stream whose base rate is 0.3143 percent, whose fraud arrives in merchant-level bursts, whose
merchants straddle every temporal boundary, and three of whose columns only exist forty-five
days after the decision they would inform; fix the operating point from the review team's
daily capacity rather than from any statistical optimum; put the course's imbalance material
to work by measuring twenty remedies — class weights, six resampling schemes, four balanced
ensembles and plain threshold selection — at that capacity-constrained point rather than at
0.5, and report which of them actually won; price resampling done before the split; repair
the prior shift every remedy introduces; build the daily queue by expected value against a
row-dependent break-even; and write the alerting plan around the fact that the only monitor
able to see a genuine change of attack arrives forty-five days late.
- Estimated duration
- 210 minutes
- Prerequisites
- chapters 026, 027, 028, 029, 034, 039, 050, 051, 052, 054, 055, 057,
059, 062, 064, 075, 076, 078, 079, 081 and 082, and labs 1 and 2
(
085.0-lab01-churn-classification.md, 086.0-lab02-medical-screening.md) - Associated quizzes
087.1-quiz-capacity-constrained-operating-points.md to
087.9-quiz-fraud-monitoring-and-drift.md