How swampybedeveloping/datathon_2026 works
This repository is an imbalanced-transaction fraud-detection project: its analysis develops and evaluates an explainable logistic-regression model, and its scoring command applies that saved model to raw CSV transactions. The principal scoring path is CSV input → feature cleaning and engineering → model scoring and threshold flags → prediction CSV. The README also documents analysis stages not present in the sampled source excerpts; those are represented at notebook level, with relationships based on README statements rather than inspected notebook code.
Main entry points: fraud_detection.ipynb Fraud analysis · predict.py CSV scoring CLI · model.pkl Model and threshold · features.py Feature engineering
Components 13
Model analysis
- Fraud analysis
notebooks/fraud_detection.ipynb - Exploratory analysis
notebooks/fraud_detection.ipynb - Leak-free preprocessing
notebooks/fraud_detection.ipynb - Feature engineering
src/fraud_detection/features.py - Model comparison
notebooks/fraud_detection.ipynb - Threshold selection
notebooks/fraud_detection.ipynb - Trust checks
notebooks/fraud_detection.ipynb
Scoring runtime
- CSV scoring CLI
src/fraud_detection/predict.py
Data and artifacts
- Training transactions
data/train.csv - Model and threshold
models/model.pkl - Test transactions
data/test.csv - Prediction CSV
outputs/test_predictions.csv
Other components
- Model analyst
Connections 14
- Model analyst → Fraud analysis: runs analysis
README.md - Training transactions → Fraud analysis: analyzes
README.md - Fraud analysis → Exploratory analysis: includes
README.md - Fraud analysis → Leak-free preprocessing: includes
README.md - Fraud analysis → Feature engineering: uses features
README.md - Fraud analysis → Model comparison: compares models
README.md - Fraud analysis → Threshold selection: selects threshold
README.md - Fraud analysis → Trust checks: evaluates trust
README.md - Fraud analysis → Model and threshold: saves model
README.md - Model analyst → CSV scoring CLI: scores CSV
README.md - Test transactions → CSV scoring CLI: provides input
README.md - CSV scoring CLI → Feature engineering: calls make_features
src/fraud_detection/predict.py - Model and threshold → CSV scoring CLI: loads model
src/fraud_detection/predict.py - CSV scoring CLI → Prediction CSV: writes predictions
src/fraud_detection/predict.py