Fraud-Scanner

Anomaly detection API for transaction scoring. Exposes a FastAPI service that returns a fraud probability and binary prediction from a JSON payload (e.g. Time, Amount, Hour).

Requirements

• Python 3.11+
• Or Docker

Setup

pip install -r requirements.txt

Run locally

uvicorn api.main:app --host 127.0.0.1 --port 8000

Then open http://127.0.0.1:8000/docs for the interactive API docs.

Run with Docker

From the project root:

docker build -t fraud-scanner .
docker run -p 8000:8000 fraud-scanner

The API is available at http://localhost:8000.

API

Method	Path	Description
GET	/health	Health check → {"status":"ok"}
POST	/predict	Score a transaction (JSON body)

Example /predict request:

{ "Time": 0.0, "Amount": 100.0, "Hour": 14 }

Example response:

{
  "prob": 0.23,
  "pred": 0,
  "threshold": 0.5,
  "missing_features": []
}

• prob: fraud probability (0–1)
• pred: 1 if fraud, 0 otherwise (based on threshold)
• missing_features: list of expected features not sent (filled with 0.0)

Tests

pytest

Model

• If models/best_model.joblib exists, it is loaded at startup (must contain model, threshold, features).
• If not, a small default logistic regression model is created and saved so the API runs out of the box. Replace models/best_model.joblib with your own trained bundle when ready.

Project layout

• api/main.py — FastAPI app and /health, /predict
• src/ — training and scoring scripts (train_rf.py, batch_score.py, etc.)
• tests/ — API and smoke tests
• notebooks/ — EDA and exploration