Outlier flags with IQR and z-scores

Flag univariate outliers per column and count them; do not delete yet. Concepts: IQR rule, z-score, robust z (median/MAD); flags vs deletion Project A: Compare IQR, z-score and modified z (your 3-method table). Do NOT apply IQR/MAD rules to discrete low-cardinality columns: num_dependents (IQR = 0 flags 155 rows) and num_existing_credits (MAD = 0 flags 367) are artefacts, not outliers. Project B: Pressure errors are real errors; heat-wave temperatures and peak loads are real events.