Sandbox Protocol_Active

ICM Fraud Detection / BATCH_007

Protocol_Brief

Challenge

How do you build a model that can catch rare fraud events without slowing down the user experience?

Solution

A combination of cost-sensitive learning (XGBoost) and a highly optimized Python inference service.

“I built this to tackle one of the hardest problems in ML: extreme class imbalance (only 1% fraud). I learned how to use XGBoost's internal weighting and feature engineering to maintain sub-100ms latency. This project was my introduction to building high-throughput inference services with FastAPI.”

Explored XGBoost's scale_pos_weight to handle extreme 30:1 class imbalance.
Built a low-latency inference pipeline optimized for sub-100ms scoring.
Learned the importance of feature parity between training and production environments.
Implemented a stateless FastAPI service designed for easy horizontal scaling.

Key_Modules

LatencySecurityStreams

Loadout_Specs

PythonFastAPIXGBoostPandasScikit-learnJoblib

Sys_Diagnostics

Node_Count:4

Fidelity:Imbalanced Learning

Matrix_State:STABLE

icm-fraud-detection.ais-cloud.dev

Viewport: Matrix_Simulated

Step 01

Imbalance Logic

Review how I configured the model to focus on rare fraud cases.