GenAI Scam Detection System: A Fraud Identification Platform Combining Generative AI and Traditional Machine Learning

GenAI Scam Detection System is a comprehensive AI-driven cybersecurity platform that integrates generative AI, natural language processing (NLP), and traditional machine learning technologies, providing real-time fraud detection and protection capabilities. The system includes a React Native mobile application, Node.js/Express backend API, Flask machine learning microservice, and Python attack simulator, covering personal to enterprise-level scenarios and capable of identifying various online threats such as phishing attacks and bank fraud.

In today's digital age, online fraud methods are becoming increasingly rampant, including phishing emails, fake OTP verification codes, counterfeit bank websites, social engineering attacks, etc. Ordinary users face unprecedented security threats. Traditional protection methods are difficult to deal with new and complex fraud patterns, so there is an urgent need for a comprehensive detection system that integrates cutting-edge technologies.

Multi-Layer Detection Mechanism

The system adopts a multi-layer detection architecture: traditional machine learning models (based on scikit-learn, TF-IDF feature extraction, logistic regression/naive Bayes) are responsible for fast initial screening; the generative AI layer (supporting Google Gemini and OpenAI API) provides deep semantic analysis and interpretable suggestions, balancing efficiency and accuracy.

Tech Stack

Layer	Tech Stack
Mobile	React Native + Expo + React Navigation + Axios
Backend API	Node.js + Express.js + MongoDB Atlas + JWT Authentication
Machine Learning	Python + Flask + scikit-learn + pandas + transformers
Generative AI	Google Gemini API / OpenAI API

API Design

The backend follows RESTful principles and provides:

• POST /auth/signup/login: User registration and JWT authentication
• POST /analyze: Full analysis (ML detection + GenAI explanation)
• POST /predict: ML prediction only
• POST /report: Fraud report submission
• GET /history: Historical record query

Mobile User Experience

The React Native application includes five modules: Home (overview and statistics), Scanner (real-time text analysis), URL Checker, History, and Profile, adopting a dark security theme and glassmorphism design.

The project includes an independent Python attack simulator that can send carefully constructed phishing payloads, used for:

• Verifying the system's ability to detect different attack variants
• Evaluating the model's performance against adversarial samples
• Continuously improving defense strategies
• Stress testing and performance benchmarking This 'defense through offense' approach ensures the system's stability when facing real threats.

Deployment

Detailed deployment documents are provided, supporting local and cloud deployment. Environment variables clearly separate sensitive information (MongoDB connection, JWT secret key, etc.), complying with security best practices.

Future Enhancements

The roadmap plans include: real-time SMS interception (Android), voice call fraud detection (Whisper+GenAI), browser extension, cross-device federated learning, and push notifications for new fraud patterns.

For ordinary users: provides active protection methods to conduct security checks before clicking/replying; for enterprises: can be integrated into customer service systems or SOC as an automated screening line. The project demonstrates the organic combination of traditional ML efficiency and generative AI's understanding ability, providing a reusable architecture template for AI application development and having wide reference value.