Skip to content

Latest commit

 

History

History
108 lines (76 loc) · 4.88 KB

README.md

File metadata and controls

108 lines (76 loc) · 4.88 KB

PayKrypt

PayKrypt is an advanced fraud detection and secure banking system designed for the financial future of Viksit Bharat 2047, implementing cutting-edge cryptography and privacy-preserving machine learning techniques.

Project Overview

This project creates a next-generation financial security infrastructure. PayKrypt combines sophisticated fraud detection capabilities with cryptographic methods to ensure long-term security and privacy in financial transactions.

Data Preprocessing

The project includes extensive data preprocessing steps for the generated data:

  • Feature engineering based on transaction time
  • Quantile and power transformations for numeric features
  • Dimensionality reduction techniques
  • Probabilistic encoding for categorical variables
  • Handling of missing values

Data Generation Using GANs

PayKrypt uses Generative Adversarial Networks (GANs) to create synthetic financial transaction data, taking the IEEE-CIS Fraud Detection dataset as a reference: IEEE-CIS Fraud Detection Competition

Key benefits of our GAN-based approach:-

  • Creates realistic transaction patterns that preserve statistical properties of real financial data
  • Generates diverse fraud scenarios for comprehensive model training
  • Enables unlimited training data without privacy concerns
  • Simulates rare fraud patterns that might be underrepresented in original datasets

Advanced Fraud Detection Methods

PayKrypt employs a multi-model ensemble approach to achieve state-of-the-art fraud detection accuracy:

Convolutional Neural Networks (CNN)

  • Temporal Pattern Recognition: Our CNN architecture treats transaction sequences as image-like structures to identify spatial-temporal fraud patterns
  • Multi-channel Feature Maps: Separate channels process different feature categories (transaction amounts, timestamps, merchant categories)
  • Hierarchical Pattern Detection: Deeper layers capture complex fraud patterns that traditional methods might miss

Long Short-Term Memory (LSTM) Networks

  • Sequential Transaction Analysis: LSTM models capture long-range dependencies in customer transaction history
  • Anomaly Detection: Identifies deviations from established customer behavior patterns
  • Temporal Fraud Signatures: Recognizes specific temporal patterns associated with fraud

Hybrid Model Architecture

  • CNN-LSTM Fusion: Combines CNN's spatial pattern recognition with LSTM's temporal analysis
  • Adaptive Feature Importance: Weights different features based on their predictive power
  • Ensemble Decision Making: Final fraud determination uses weighted outputs from multiple models
  • Continuous Learning: Models update in real-time as new patterns emerge while preserving privacy

AI Banking Assistant

PayKrypt includes an intelligent AI chatbot designed to assist users with banking-related queries and provide support during fraudulent situations:

  • Fraud Response Guidance: Provides step-by-step assistance to users who have experienced fraudulent transactions
  • Personalized Support: Tailors advice based on the specific type of fraud encountered and the user's financial situation
  • Real-time Assistance: Available 24/7 to help users immediately after detecting suspicious activity

Homomorphic Encryption

PayKrypt implements encryption schemes for homomorphic encryption:

  • Secure Homomorphic Encryption: Our system utilizes encryption schemes that protect sensitive data
  • Homomorphic Computation: Enables computations on encrypted data without decryption
  • Future-Proof Architecture: Designed to adapt to evolving cybersecurity challenges

Federated Learning System

PayKrypt implements a federated learning architecture with secure encryption:

  • Decentralized Model Training: Each financial institution trains models locally on their own data
  • Secure Parameter Sharing: Model updates are protected with homomorphic encryption
  • Zero-Knowledge Proofs: Verification of data integrity without revealing sensitive information
  • Secure Aggregation: A central server aggregates encrypted model parameters without compromising security

This approach creates a nationwide secure financial network aligned with Viksit Bharat 2047's vision of technological sovereignty and advanced digital infrastructure.

Installation

# Clone the repository
git clone https://github.com/KrishBnsl/paykrypt.git
cd paykrypt

# Create a virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Usage

  1. Data generation and preprocessing:
jupyter notebook data_preprocess.ipynb
  1. Run the federated learning system with encryption:
jupyter notebook privacy_federated_learning.ipynb

Team: Null (N) Void

  • Krish Bansal
  • Mayank Jangid
  • Daksh Gangwar
  • Kkartik Aggarwal