Skip to main content
Get your fraud detection system up and running in just a few steps.

​
Prerequisites

Before you begin, ensure you have the following installed:
  • Python 3.8 - The project requires Python 3.8 for compatibility with specific library versions
  • Miniconda or Anaconda - For creating isolated Python environments
  • Git - For cloning the repository

​
Setup Steps

1

Clone the repository

Clone the fraud detection project to your local machine:
git clone https://github.com/sujith52/fraud.git
cd fraud
2

Create conda environment

Create a new conda environment with Python 3.8:
conda create -n fraud-env python=3.8 -y
conda activate fraud-env
Using a dedicated environment ensures dependency isolation and prevents conflicts with other Python projects.
3

Install dependencies

Install all required packages from requirements.txt:
pip install -r requirements.txt
This will install:
  • Flask 1.1.1 (web framework)
  • scikit-learn 0.22.1 (ML algorithms)
  • XGBoost 0.90 (gradient boosting)
  • pandas 0.25.3 (data processing)
  • Flask-MonitoringDashboard 3.0.6 (monitoring)
  • And 40+ other dependencies
4

Run the application

Start the Flask application:
python main.py
The server will start at http://127.0.0.1:5001
* Serving Flask app 'main' (lazy loading)
* Environment: production
* Debug mode: on
* Running on http://127.0.0.1:5001/ (Press CTRL+C to quit)

​
Train a Model

Before making predictions, you need to train a model on your insurance claims data.

​
Using the API

Send a POST request to the /train endpoint: 
curl -X POST http://127.0.0.1:5001/train \
  -H "Content-Type: application/json" \
  -d '{"folderPath": "Training_Batch_Files/"}'
Training may take several minutes depending on the dataset size. The system will:
  1. Validate and preprocess the data
  2. Perform K-Means clustering
  3. Train XGBoost and SVM models per cluster
  4. Select and save the best model for each cluster

​
Make a Prediction

Once models are trained, you can predict fraud on new insurance claims.

​
Using the API

Send a POST request to the /predict endpoint: 
curl -X POST http://127.0.0.1:5001/predict \
  -H "Content-Type: application/json" \
  -d '{"filepath": "Prediction_Batch_files/"}'

​
Response

Prediction File created at Prediction_Output_File/Predictions.csv!!!

​
View Results

The prediction results will be saved as a CSV file at Prediction_Output_File/Predictions.csv: 
Predictions
N
Y
N
N
Y
Where:
  • Y = Fraud detected
  • N = No fraud detected

​
Access Monitoring Dashboard

View API performance metrics and usage statistics: 
http://127.0.0.1:5001/dashboard
The dashboard provides:
  • Request/response times
  • Endpoint usage statistics
  • Error rates
  • Performance graphs
The monitoring dashboard uses Flask-MonitoringDashboard and stores metrics in flask_monitoringdashboard.db.

​
Next Steps

Installation Guide

Detailed installation instructions and troubleshooting

Training Guide

Learn about the model training pipeline

API Reference

Complete API documentation

Deployment

Deploy to production with Gunicorn

​
Troubleshooting

Change the port by setting the PORT environment variable:
PORT=8000 python main.py
Ensure you’ve activated the conda environment and installed all dependencies:
conda activate fraud-env
pip install -r requirements.txt
Ensure your training data CSV files are in the Training_Batch_Files/ directory with the correct naming format: fraudDetection_[DATESTAMP]_[TIMESTAMP].csv

Build docs developers (and LLMs) love

Get started for freeTalk to us