I built a fraud‑risk detection agent reducing expected loss from ₹936 to ₹88 per transaction by cost‑tuned threshold.
I am a final‑year ML student aiming to break into AI growth, and I tuned the model’s decision threshold based on monetary cost rather than standard accuracy metrics to directly minimize financial loss.
- Expected loss per transaction before cost‑tuned threshold: ₹936
- Expected loss per transaction after cost‑tuned threshold: ₹88