Start-to-production implementation of bank fraud detection using XGBoost
Most of the time, decision tree based approaches work better with structured, tabular data. More importantly, compared to deep learning approaches, they are better fit for imbalanced data like fraud data.
XGBoost is the go-to library when it comes to implementing gradient boosting
algorithms
With uv installed
## 1. Clone the repo
git clone [email protected]:hmzdot/fraud-ml.git
cd fraud-ml
## 2. Install dependencies
# Either with uv
uv sync
# Or with pip (assuming you've setup venv yourself)
pip3 install .
## 3. Train the model
# Either with uv
uv run src/train.py
# Or with python3 (assuming venv is active)
python3 src/train.py
## 4. Evaluate another dataset
# Either with uv
uv run src/eval.py ./snapshots/{model} {path/to/data}
# Or with python3
python3 src/eval.py ./snapshots/{model} {path/to/data}