How tejabdhhskas-a11y/fraud-detection-system works
This repository is a UPI fraud-detection system: synthetic transaction data is generated, used to train and evaluate an XGBoost model, and the saved model is served through a FastAPI prediction endpoint. A separate SHAP workflow explains model behavior. The main request path accepts transaction features, scales them, predicts fraud probability, and returns a risk classification. The README documents the overall workflow and reported model performance; the graph’s implementation links are based on the sampled source, so unsampled behavior is not inferred.
Main entry points: generate_data.py Transaction Dataset · train_model.py Training Pipeline · app.py FastAPI Service · explain.py SHAP Analysis
Components 11
Synthetic Data
- Transaction Generator
generate_data.py - Transaction Dataset
generate_data.py
Model Training
- Training Pipeline
train_model.py - Fraud Model
fraud_model.pkl - Feature Scaler
scaler.pkl
Prediction API
- FastAPI Service
app.py - Feature Scaling
app.py - Fraud Inference
app.py - Risk Response
app.py
Model Explanations
- SHAP Analysis
explain.py
Other components
- API Client
Connections 13
- API Client → FastAPI Service: submits transaction
README.md - Transaction Generator → Transaction Dataset: generates
generate_data.py - Transaction Dataset → Training Pipeline: loads CSV
train_model.py - Training Pipeline → Fraud Model: saves model
train_model.py - Training Pipeline → Feature Scaler: saves scaler
train_model.py - FastAPI Service → Feature Scaling: builds features
app.py - Feature Scaling → Feature Scaler: transforms with
app.py - FastAPI Service → Fraud Inference: requests prediction
app.py - Fraud Inference → Fraud Model: predicts with
app.py - Fraud Inference → Risk Response: returns risk
app.py - Transaction Dataset → SHAP Analysis: samples transactions
explain.py - SHAP Analysis → Feature Scaler: transforms sample
explain.py - SHAP Analysis → Fraud Model: explains predictions
explain.py