Volume 17, Number 3

Deep Learning Solutions for Source Code Vulnerability Detection

  Authors

Nin Ho Le Viet, Long Phan, Hieu Ngo Van and Tin Trinh Quang, Duy Tan University, Vietnam

  Abstract

Detecting vulnerabilities in software source code has become a critical aspect of developing secure systems. Traditional methods are increasingly limited in processing complex code structures and generalizing to previously unseen scenarios. In response, advanced deep learning models such as CNN, LSTM, Bi-LSTM, Self-Supervised Learning (SSL), and Transformer have demonstrated potential in automatically capturing the semantic and contextual characteristics embedded in code. This paper serves as both a guided review and a quantitative comparison of the performance of deep learning models for vulnerability detection. Key evaluation indicators, such as accuracy, F1-score, and computational cost, are used to benchmark the models. Results highlight that Transformer achieves the highest accuracy (96.8%), while CNN remains favorable in low-resource environments. The paper concludes with model selection guidelines and suggestions for future enhancements in real-world deployment.

  Keywords

Source Code Vulnerability, Deep Learning, CodeBERT, GraphCodeBERT, GPT-4.

Full Text    Volume 17

 

 

© AIRCC Publishing Corporation