Journal entry anomaly detection for audit

The brief

Auditors cannot read every entry, so they sample. Build a system that scores every journal entry for risk using rules (round numbers, weekends, rare accounts, near-limit amounts) and a model, and presents the top entries with reasons so the auditor samples smartly. Good looks like: planted anomalies found at a high rate, few false alarms, and each flag explained in audit language.

What you will build

1. Generate a year of journal entries
2. Plant anomalies of several types
3. Build rule-based audit tests
4. Add an isolation forest or similar model
5. Combine into a risk score with reasons
6. Measure detection and false alarms

Free, open data to build it with