Fraud Detection System for Financial Transactions

Dataset Link: Kaggle

Reference: GitHub

GitHub Repository: Fraud Detection System

Problem Statement

Financial institutions face challenges in detecting and preventing fraudulent transactions. UPI-based financial fraud leads to significant monetary losses.

Project Overview

This project aims to detect fraudulent financial transactions in real time using machine learning. It leverages AI/ML models to analyze transaction patterns, ensuring scalability and adaptability to emerging fraud tactics. A web application built with Flask provides an interface for users to input transaction details and receive fraud risk predictions.

Objective & Solution Approach

• Develop an AI/ML model to analyze transaction patterns and detect fraud in real time.
• Ensure scalability for handling large transaction volumes.
• Implement adaptive learning for detecting emerging fraud patterns.

Technology & Models Used

Machine Learning Models:

• Logistic Regression
• Decision Tree
• K-Nearest Neighbors (KNN)
• Random Forest
• Naïve Bayes
• XGBoost (Best Performing)

Technology Stack:

• Python
• Flask (Web App)
• Pandas
• Scikit-learn
• XGBoost

Implementation

Dataset Features:

• Transaction type, amount, source & destination balances, timestamp

Preprocessing:

• Feature engineering, normalization using scaler.pkl

Model Training & Evaluation:

• Trained on historical transaction data
• Best accuracy achieved using XGBoost

Installation

1. Clone the repository:

   git clone https://github.com/PRIYAtechky/Fraud-Detection-System-for-Financial-Transaction.git

2. Navigate to the project directory:

   cd Fraud-Detection-System-for-Financial-Transaction

3. Install dependencies:

   pip install -r requirements.txt

How to Use

1. Update File Paths:
   • Open app.py and update the file paths for scaler.pkl, xgb_model.pkl, and column_names.pkl to match your system's directory structure.

   with open(r'path/to/scaler.pkl', 'rb') as file:
       scaler = pickle.load(file)
   
   with open(r'path/to/xgb_model.pkl', 'rb') as file:
       model = pickle.load(file)
   
   with open(r'path/to/column_names.pkl', 'rb') as file:
       column_names = pickle.load(file)

2. Run the Flask application:

   python app.py

3. Access the Web Application:
   • Once app.py runs successfully, it will display a local server address in the output.
   • Open your browser and enter the URL provided in the terminal (default: http://127.0.0.1:5000/).
4. Input transaction details to check for fraud risk.