Advancing Financial Security - A Hybrid Fusion of Neural Network & Rule-Based System

Published in: Journal of Information and Optimization Sciences (JIOS), Taru Publications
DOI: 10.47974/JIOS-1932
Index: Web of Science

Overview

This project implements a hybrid fraud detection system that integrates:

• Neural Network for learning fraud patterns
• Rule-Based System for improved interpretability & accuracy

The hybrid system improves accuracy to 98.41%, outperforming traditional models by leveraging both AI-based detection and human-understandable rules.

Dataset Used

• IEEE-CIS 2019 Credit Card Fraud Detection Dataset
• Available on Kaggle: 🔗 IEEE Fraud Detection Dataset

How It Works

1. Model Training (training.py)
   • The deep learning model is trained on transaction data.
   • Techniques used: Log transformation, focal loss, Nadam optimizer.
   • The trained model detects fraudulent transactions based on patterns in the data.
2. Hybrid Integration (hybrid.py)
   • The trained model’s predictions are passed to the rule-based system.
   • The system applies expert-defined fraud detection rules to refine results.
   • This improves interpretability and reduces false positives.

Features

• Neural Network Training: Uses log transformation, focal loss, and Nadam optimizer
• Rule-Based System: Applies expert-defined fraud detection rules
• Hybrid Approach: Improves model performance & interpretability
• Optimized for Real-World Fraud Detection

Technologies Used

• Python 3.12.9
• TensorFlow/Keras
• Scikit-Learn
• Pandas & NumPy
• Matplotlib & Seaborn

Installation Guide

1. Clone the Repository

    git clone https://github.com/Suryatejaeasari/Hybrid-Credit-Card-Fraud-Detection.git
    cd Hybrid-Credit-Card-Fraud-Detection
   

2. Create a Virtual Environment (Optional but Recommended)

   python -m venv venv
   source venv/bin/activate  # For macOS/Linux
   venv\Scripts\activate     # For Windows
   

3. Install Dependencies

    pip install -r requirements.txt
   

4. Run the Project

   python training.py  # Train the model
   python hybrid.py  # Apply rule-based system