5/ Wrapped it in a production-style Patrol class with a check_transaction() method — so it's not just a notebook, it's something that could actually run.
Full writeup + code: https://t.co/axtqdBbl8P
Open to remote Data Analyst roles 👊
4/ Tuned the decision threshold to balance precision and recall.
Catching more fraud (recall) usually means more false alarms (precision drops). Landed at ~74.7% recall — a deliberate tradeoff, not a default.
6 months ago I couldn't write a SELECT statement.
Now I've shipped 6 real data projects — including a fraud detection pipeline modeling risk across 1.56M transactions.
Self-taught. No bootcamp. No CS degree.
Portfolio: https://t.co/axtqdBbl8P
Open to remote Data Analyst roles