How fan000n/fraud-detection-ml works
This repository is a notebook-led fraud-detection ML project, not a deployed transaction service. Its user-to-result workflow runs from raw-data inspection through leakage-conscious preprocessing and model comparison to threshold selection and final test-set evaluation. The README documents the notebook order and preprocessing behavior; the source excerpts show the reusable preprocessing, model-builder, and metric modules. Notebook-to-module wiring is not shown, so those links are left unverified rather than asserted.
Main entry points: 01_eda.ipynb Explore data · 02_modeling.ipynb Compare models · 03_threshold_and_final_evaluation.ipynb Tune threshold · data Fraud CSV data
Components 9
Notebook workflow
- Explore data
notebooks/01_eda.ipynb - Compare models
notebooks/02_modeling.ipynb - Tune threshold
notebooks/03_threshold_and_final_evaluation.ipynb
Reusable ML code
- Prepare features
src/preprocessing.py - Model builders
src/train.py - Metric evaluation
src/evaluate.py
Data and outcome
- Fraud CSV data
data - Fraud metrics
Other components
- Project analyst
Connections 8
- Project analyst → Explore data: runs first
README.md - Fraud CSV data → Explore data: inspects
README.md - Explore data → Compare models: continues to
README.md - Compare models → Tune threshold: continues to
README.md - Explore data → Prepare features: uses pipeline
README.md - Compare models → Model builders: uses builders
- Tune threshold → Metric evaluation: uses metrics
- Tune threshold → Fraud metrics: reports
README.md