🔍 GraphCodeBERT Vulnerability Detector

An AI-powered system that detects logic-flaw vulnerabilities in C/C++ code using CodeBERT, code slicing, and graph-based analysis (AUG-PDG).

---

🚀 Overview

This project analyzes C/C++ source code to identify potential vulnerabilities without executing it.
It focuses on risky patterns, builds dependency graphs, and uses a fine-tuned transformer model to classify code as:

• ⚠️ Vulnerable
• ✅ Safe

---

⚙️ Key Features

• 🔹 Code Slicing – Extracts only risk-prone parts of code (pointers, memory ops, loops)
• 🔹 Graph-Based Analysis – Builds a simplified Program Dependency Graph (PDG)
• 🔹 Transformer Model – Uses microsoft/codebert-base for classification
• 🔹 End-to-End Pipeline – Input raw C/C++ → Output prediction + graph
• 🔹 Visualization – Displays dependency graph using NetworkX

---

🧠 Methodology

1. Input Code
2. Slicing
   • Extract risky lines (e.g., malloc, strcpy, pointers, loops)
3. Graph Construction
   • Nodes → code lines
   • Edges → control flow + data flow
4. Model Inference
   • Tokenize sliced code
   • Pass through CodeBERT
   • Predict vulnerability
5. Output
   • Vulnerability label
   • Graph visualization

---

📊 Results

Metric	Score
Accuracy	74%
Precision	0.696
Recall	0.772
F1 Score	0.732

The model prioritizes recall, making it effective for detecting most vulnerabilities.

---

🛠️ Tech Stack

• Python 🐍
• PyTorch 🔥
• HuggingFace Transformers 🤗
• NetworkX 🕸️
• Matplotlib 📊
• Scikit-learn

---

📂 Project Structure

GraphCodeBERT-VulnDetector ┣ src/ ┃ ┣ model.py ┃ ┣ graph_extractor.py ┃ ┣ infer.py ┣ notebooks/ ┣ data/ ┣ requirements.txt ┣ README.md

---

▶️ How to Run

1. Install dependencies

pip install -r requirements.txt


from infer import predict_code

code = """
int main() {
    char buffer[10];
    gets(buffer);
    return 0;
}
"""

result, graph = predict_code(code)
print(result)

import matplotlib.pyplot as plt
import networkx as nx

nx.draw(graph, with_labels=True)
plt.show()