🛡️ DefectPrediction

AI-Powered Vulnerability Detection for C/C++ Code

DefectPrediction is a full-stack machine learning web application that detects security vulnerabilities in C/C++ source code. It combines manual safety rules, static heuristics, and deep learning inference using a fine-tuned GraphCodeBERT model to reduce false positives and improve reliability.

🧠 Architecture Overview

🎨 Application Interface

---

🚀 Features

• 🔍 Detects common vulnerabilities:
  • Buffer overflows
  • Null pointer dereferences
  • Use-after-free
  • Out-of-bounds access
• 🧠 Fine-tuned GraphCodeBERT model
• 🛑 Manual rule-based checker (highest priority)
• 🟢 Structural clean overrides to reduce false positives
• 🌐 Full-stack application:
  • FastAPI backend
  • React + Vite frontend
• 📊 Confidence-based predictions
• 🎨 Modern UI with integrated code editor

---

🧩 Project Structure

DefectPrediction/
├── Backend/
│   ├── app.py
│   ├── requirements.txt
│   └── final_graphcodebert_balanced_best/   (download separately)
├── FrontEnd/
│   ├── src/
│   ├── package.json
│   └── vite.config.ts
├── .gitignore
├── README.md

---

🧠 Model Information

• Model: GraphCodeBERT
• Dataset: CodeXGLUE Devign
• Task: Binary classification (Clean vs Defective)
• Frameworks: PyTorch, HuggingFace Transformers

⚠️ The trained model is not included in this repository due to size constraints.

---

📥 Model Download (Required)

🔗 Download the trained model from Google Drive

👉 https://drive.google.com/file/d/1Q3_x5eaYQ-jlntAgGuo9sau5Kic-P5r8/view

📂 Setup Steps

1. Download and extract:

   final_graphcodebert_balanced_best/
   

2. Place it inside:

   Backend/final_graphcodebert_balanced_best/
   

3. Ensure the folder contains:

   • config.json
   • model.safetensors
   • tokenizer.json
   • tokenizer_config.json

---

⚙️ Backend Setup (FastAPI)

cd Backend
python -m venv venv
source venv/bin/activate   # macOS / Linux
venv\Scripts\activate    # Windows
pip install -r requirements.txt
uvicorn app:app --reload

Backend URL: http://127.0.0.1:8000
API Docs: http://127.0.0.1:8000/docs

---

🎨 Frontend Setup (React + Vite)

cd FrontEnd
npm install
npm run dev

Frontend URL: http://localhost:5173

---

🧪 How It Works

Safe API Override
    ↓
Manual Defect Rules
    ↓
Static Safety Heuristics
    ↓
GraphCodeBERT Model (60% confidence threshold)

---

🧠 Decision Strategy

• Manual rules override everything (guaranteed defects)
• Structural clean overrides suppress ML false positives
• ML inference is used only when rules are inconclusive
• Code is marked defective only if confidence ≥ 60%

---

🛠️ Technologies Used

• Backend: FastAPI, PyTorch, HuggingFace Transformers
• Frontend: React, Vite, Tailwind CSS
• Model: GraphCodeBERT
• Languages: Python, TypeScript, C/C++

---

📌 Notes

• The system prioritizes false-negative avoidance
• Pointer-heavy clean code may still receive low defect confidence
• Rule-based overrides improve real-world usability

---

📜 License

This project is intended for educational and research purposes.

---