Skip to content

Latest commit

 

History

5 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🛡️ FraudShield Intelligence

Enterprise-Grade Real-Time Financial Fraud Detection Platform

Python 3.9+ FastAPI Streamlit XGBoost Docker

FraudShield Intelligence is an end-to-end machine learning platform engineered to detect fraudulent financial transactions with ultra-low latency. By pairing an optimized extreme gradient boosting framework with a scalable asynchronous REST API and an intuitive analytical dashboard, the platform delivers real-time risk scoring alongside model explainability powered by Explainable AI (XAI) paradigms.


Live Demo

Service URL
Application https://fraudshield-intelligence-evci2hh3g7wzrytqmg848y.streamlit.app

Overview

FraudShield Intelligence is a production-ready machine learning system designed to identify potentially fraudulent financial transactions in real time. The project goes beyond a single model — it systematically trains and benchmarks three classifiers (Logistic Regression, Random Forest, and XGBoost) on a 6.3M+ transaction dataset, with XGBoost selected as the final model based on superior recall and F1 performance on the highly imbalanced fraud detection task.

The platform combines a high-performance classification backend with a modern web interface and REST API, enabling users to evaluate transaction risk through an intuitive dashboard — with SHAP-powered explanations for every prediction.


Application Preview

Home Page

Screenshot 2026-06-29 at 2 49 45 AM

Prediction Result

Screenshot 2026-06-29 at 2 51 08 AM

SHAP Explainability

Screenshot 2026-06-29 at 2 53 18 AM

Features

  • Real-time fraud prediction via REST API
  • Rigorous multi-model training and evaluation pipeline
  • XGBoost classifier selected through benchmarked comparison
  • Interactive Streamlit frontend dashboard
  • FastAPI backend with Swagger UI documentation
  • SHAP Explainable AI — per-prediction feature attribution
  • Docker and Docker Compose support
  • Cloud deployment (Render + Streamlit Community Cloud)
  • Confidence score visualization
  • Transaction summary dashboard

Model Selection & Empirical analysis

Three classifiers were trained and rigorously evaluated to identify the best performer for fraud detection — where recall is the critical metric due to the severe class imbalance inherent in financial fraud datasets.

Model Comparison

Metric Logistic Regression Random Forest XGBoost
Accuracy Baseline 99.97% 99.98%
Precision Baseline 98.02% 96.60%
Recall Baseline 78.59% 85.98%
F1 Score Baseline 87.24% 90.98%
False Negatives High 446 292

XGBoost — Selected Model

Metric Score
Accuracy 99.978%
Precision 96.60%
Recall 85.98%
F1 Score 90.98%
Confusion Matrix:
                   Predicted Legitimate   Predicted Fraud
Actual Legitimate        1,588,509               63
Actual Fraud                   292            1,791

XGBoost was selected for its superior recall and F1 score — catching the highest proportion of actual fraudulent transactions while maintaining strong precision. Its gradient boosting framework handles class imbalance more effectively, delivering a 7.4-point recall improvement over Random Forest and catching 154 additional fraudulent transactions per evaluation cycle.


System Architecture

           User
             │
             ▼
  Streamlit Frontend
  (Streamlit Community Cloud)
             │
       REST API Call
             │
             ▼
    FastAPI Backend
       (Render)
             │
             ▼
  XGBoost Fraud Detection Model
       + SHAP Explainer
             │
             ▼
  Fraud Prediction + Feature Attributions

Tech Stack

Category Technology
Language Python
Machine Learning XGBoost, Scikit-learn
Backend FastAPI
Frontend Streamlit
Explainability SHAP
API Documentation Swagger UI
Containerization Docker & Docker Compose
Cloud Deployment Render (backend), Streamlit Community Cloud (frontend)
Version Control Git & GitHub

Input Features

Feature Description
Step Hours elapsed since first transaction
Transaction Type CASH_IN, CASH_OUT, PAYMENT, DEBIT, TRANSFER
Amount Transaction amount (USD)
Sender Balance Before Sender's account balance before transaction
Sender Balance After Sender's account balance after transaction
Receiver Balance Before Receiver's account balance before transaction
Receiver Balance After Receiver's account balance after transaction
Flagged Fraud Rule-based fraud indicator from source system

Project Structure

FraudShield-Intelligence/
│
├── backend/
│   ├── app.py                          # FastAPI application & prediction endpoints
│   ├── credit_fraud_xgb.pkl            # Trained XGBoost model artifact
│   ├── requirements.txt
│   └── Dockerfile
│
├── frontend/
│   ├── streamlit_app.py                # Streamlit UI & SHAP visualization
│   ├── requirements.txt
│   └── Dockerfile
│
├── notebook/
│   └── Credit_Card_Fraud_Detection.ipynb   # Full training & benchmarking pipeline
│
├── dataset/
│   └── README.md                       # Dataset download instructions
│
├── docker-compose.yml
└── README.md

Running Locally

Clone the repository

git clone https://github.com/eklavya072/FraudShield-Intelligence.git
cd FraudShield-Intelligence

Backend

cd backend
pip install -r requirements.txt
uvicorn app:app --reload

Frontend (in a separate terminal)

cd frontend
pip install -r requirements.txt
streamlit run streamlit_app.py

Docker

Run the full stack with a single command:

docker compose up --build

Dataset

This project uses the PaySim synthetic financial transaction dataset — a simulation of mobile money transactions designed for fraud detection research.

Due to GitHub file size limitations, the dataset is not included in this repository. Download it from Kaggle and place it inside the dataset/ directory before running the training notebook.


Explainable AI with SHAP

FraudShield Intelligence integrates SHAP (SHapley Additive exPlanations) to provide per-prediction transparency. For every transaction evaluated, SHAP computes the contribution of each individual feature to the final prediction — making the model auditable and interpretable rather than a black box.

This is particularly important in financial applications, where regulators and end users require justification for risk classifications.


About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages