A Survey on Software Vulnerability Detection Techniques Using Static and Dynamic Analysis

Abstract

Software vulnerabilities pose significant security risks to modern applications, particularly as software systems become increasingly complex, interconnected, and widely deployed across critical domains. Effective vulnerability detection is therefore essential for identifying security weaknesses before they can be exploited. This survey provides a comprehensive review of software vulnerability detection techniques, with particular emphasis on static analysis, dynamic analysis, and hybrid approaches. Static analysis enables early detection and broad code coverage, while dynamic analysis evaluates runtime behavior and supports vulnerability validation. Hybrid approaches combine these complementary capabilities to improve detection coverage and reduce false-positive findings. The survey further examines recent AI-driven approaches, including machine learning, deep learning, graph neural networks, transformer-based models, Large Language Models (LLMs), and multimodal learning. The reviewed studies indicate that AI-assisted and hybrid approaches can improve vulnerability detection capability, semantic understanding, vulnerability coverage, and automation compared with isolated traditional techniques. However, challenges involving false positives, dataset quality, computational cost, explainability, and model generalization remain. Overall, the findings highlight the potential of integrating traditional program analysis with AI-driven methods to develop more accurate, scalable, and reliable software vulnerability detection systems.