TranSec: Real-Time Fraud Detection

Hackathon Project: An AI-based system to detect and block financial fraud in real-time by analyzing transaction patterns.

Problem Statement: Design an AI system for real-time fraud detection using transaction patterns, ensuring secure financial operations.

GitHub: foskey51/fintech-404

Screenshots

Image

Overview

TranSec uses a LightGBM model to predict fraud in financial transactions, providing real-time insights with SHAP explanations. The frontend, built with Next.js, displays a dashboard with transaction analytics, while the Flask backend serves predictions via a /predict endpoint. Trained on the Kaggle Fraudulent Transactions Dataset.

Features

• Real-Time Detection: Classifies transactions as Fraud, Legitimate, or Manual Review.
• Explainability: SHAP-based feature importance for each prediction.
• Interactive Dashboard: Visualizes fraud metrics with a bar chart and transaction table.
• Scalable Backend: Flask API with Docker support.

Dataset

The model is trained on the Kaggle Fraudulent Transactions Dataset:

• Columns:
  • step: Unit of time (1 step = 1 hour, 744 steps = 30 days).
  • type: CASH-IN, CASH-OUT, DEBIT, PAYMENT, TRANSFER.
  • amount: Transaction amount in local currency.
  • nameOrig: Originating customer (dropped during training/prediction).
  • oldbalanceOrg: Sender’s balance before transaction.
  • newbalanceOrig: Sender’s balance after transaction.
  • nameDest: Recipient customer (dropped during training/prediction).
  • oldbalanceDest: Recipient’s balance before transaction.
  • newbalanceDest: Recipient’s balance after transaction.
• Note: nameOrig and nameDest are optional and excluded from model training and predictions.

Features

• Fraud detection using machine learning on a Kaggle dataset
• API endpoint for predicting fraudulent transactions
• Frontend built with Next.js, Framer Motion, and Chart.js
• Backend built with Flask, LightGBM, SHAP, and Pandas

Technologies Used

• Python 3.8+
• Flask
• LightGBM
• SHAP
• Pandas
• Framer Motion
• Chart.js
• Tailwind CSS
• Vite

Setup

Frontend

1. Install dependencies:

npm install

2. Run the server:

npm run dev

Backend

Linux/Mac (via Terminal)

1. Open a terminal.
2. Create a new virtual environment named fraud-404-env.
3. Activate the environment using:

   source fraud-404-env/bin/activate  # Linux/Mac
   .fraud-404-env\Scripts\activate  # Windows
   

Windows (via Command Prompt)

1. Open a command prompt.
2. Create a new virtual environment named fraud-404-env.
3. Activate the environment using:

   \fraud-404-env\Scripts\activate
   

Installing Dependencies

1. Install required dependencies by running:

   bash
   pip install -r requirements.txt
   

API Endpoint

• URL: http://localhost:5000/predict
• Method: POST
• Example request body:

[
  {
    "step": 1,
    "type": "TRANSFER",
    "amount": 1000,
    "oldbalanceOrg": 5000,
    "newbalanceOrig": 4000,
    "oldbalanceDest": 1000,
    "newbalanceDest": 2000
  }
]

Response:

[
  {
    "explanation": [
      {"feature": "newbalanceOrig", "impact": -7.5938},
      {"feature": "type", "impact": 1.1688},
      {"feature": "oldbalanceOrg", "impact": -1.1001}
    ],
    "fraud_probability": 0.04,
    "prediction": 0,
    "status": "Legitimate"
  }
]