Types of anomalies and the 'model normal' framing

Write one point, contextual and collective anomaly example per project; set the review budget (alerts per day). Concepts: Point / contextual / collective anomalies; scores vs alerts vs incidents; review capacity Project A: ULB card transactions (284,807 rows, 0.17% fraud). Labels HIDDEN during training, used only to evaluate. Point anomaly: unusual amount/feature combination. Project B: Hourly load from the EDA dataset. Contextual: normal MW for a weekday but abnormal for a holiday at 03:00; collective: a flat-lined day (meter outage).