Build and evaluate imbalanced binary classifier

You are given a binary classification dataset with severe class imbalance (positive rate ≈1%). Each row has: id, event_date (YYYY-MM-DD), categorical: region ∈ {NA, EU, APAC, LATAM, MEA}, and numerical features f1…f50. Labels are y ∈ {0,1}. Tasks: a) Build a reproducible training pipeline that: - splits temporally into train (≤2025-06-01), validation (2025-06-02–2025-08-01), and test (2025-08-02–2025-09-01); - applies standardization to numeric features and one-hot encoding to region; - handles imbalance inside CV folds (e.g., class_weight='balanced', or SMOTE within each fold without leaking validation data); - trains a strong baseline (e.g., calibrated logistic regression or gradient boosting) and outputs well-calibrated probabilities (Platt or isotonic on validation). b) Report ROC-AUC and PR-AUC on the test split; also report recall at 5% FPR and the decision threshold that maximizes F1 under the constraint recall ≥ 0.90. c) Describe how you would choose the operating point for a production system with a hard requirement of at most 2 false positives per 1,000 predictions. d) Discuss how calibration might drift over time and one technique to monitor and re-calibrate without label leakage.