Fraud Detection Model API

Project Overview

This project implements a complete pipeline for fraud detection in an insurance dataset containing categorical features. The model uses Weight of Evidence (WOE) transformation and a logistic regression scorecard for prediction.

Features

• Data preprocessing: duplicate removal, categorical validation, missing values handling.
• Model training with WOE transformation and logistic regression.
• Scorecard generation showing points per category.
• REST API to compute risk scores for clients.

Requirements

All dependencies are listed in requirements.txt.
Install them with:

pip install -r requirements.txt

Running the API

To start the API server using Uvicorn:

uvicorn main:app --reload --host 0.0.0.0 --port 8000

Once started, open in your browser: Locally: http://127.0.0.1:8000/docs From another device on the same network: http://<your_local_ip>:8000/docs

API Endpoints

POST /score

Predicts the fraud risk score for a given client.

Request body (JSON example):

{
    "Month": "Jan",
    "Make": "Pontiac",
    "AccidentArea": "Urban",
    "DayOfWeekClaimed": "Monday",
    "MonthClaimed": "Jan",
    "WeekOfMonthClaimed": "4",
    "MaritalStatus": "Married",
    "VehiclePrice": "less than 20000",
    "AgeOfVehicle": "more than 7",
    "AgeOfPolicyHolder": "41 to 50",
    "AgentType": "External",
    "AddressChange_Claim": "no change",
    "NumberOfCars": "1 vehicle"
}

Example request with cURL:

curl -X POST "http://0.0.0.0:8000/score" -H "Content-Type: application/json" -d '{
    "Month": "Jan",
    "Make": "Pontiac",
    "AccidentArea": "Urban",
    "DayOfWeekClaimed": "Monday",
    "MonthClaimed": "Jan",
    "WeekOfMonthClaimed": "4",
    "MaritalStatus": "Married",
    "VehiclePrice": "less than 20000",
    "AgeOfVehicle": "more than 7",
    "AgeOfPolicyHolder": "41 to 50",
    "AgentType": "External",
    "AddressChange_Claim": "no change",
    "NumberOfCars": "1 vehicle"
}'

Example response:

{
  "score": 466,
  "probability_fraud": 0.676999,
  "log_odds": 0.740013,
  "used_variables": [
    "Month",
    "Make",
    "AccidentArea",
    "DayOfWeekClaimed",
    "MonthClaimed",
    "WeekOfMonthClaimed",
    "MaritalStatus",
    "VehiclePrice",
    "AgeOfVehicle",
    "AgeOfPolicyHolder",
    "AgentType",
    "AddressChange_Claim",
    "NumberOfCars"
  ],
  "inputs_used": {
    "Month": "Jan",
    "Make": "Pontiac",
    "AccidentArea": "Urban",
    "DayOfWeekClaimed": "Monday",
    "MonthClaimed": "Jan",
    "WeekOfMonthClaimed": "4",
    "MaritalStatus": "Married",
    "VehiclePrice": "less than 20000",
    "AgeOfVehicle": "more than 7",
    "AgeOfPolicyHolder": "41 to 50",
    "AgentType": "External",
    "AddressChange_Claim": "no change",
    "NumberOfCars": "1 vehicle"
  }
}

Notes

• Ensure that the model artifacts (selected_vars.json, woe_mappings.json, model.pkl, etc.) are located in the artifacts/ folder before starting the API.
• Modify main.py if the model or preprocessing changes.

Results Presentation

The complete results analysis, including visualizations, metrics, and scorecard interpretation, is available in the Jupyter notebook fraud_detection.ipynb.