GRAPH-BASED MACHINE LEARNING FOR SOURCE CODE VULNERABILITY DETECTION AND SECURE SOFTWARE TESTING

Abstract

The increasing complexity of software applications and the rapid adoption of interconnected digital systems have created significant challenges in identifying source code vulnerabilities effectively. Traditional vulnerability detection techniques based on static analysis and predefined security rules often fail to capture complex relationships between software components, dependencies, and execution behaviors. This paper proposes a Graph-Based Machine Learning Framework for Source Code Vulnerability Detection and Secure Software Testing by integrating graph representation learning, machine learning, software dependency analysis, and automated security testing. The proposed framework converts source code into graph-based representations by modeling functions, variables, control flows, data dependencies, and program interactions. Graphbased learning models analyze these representations to identify vulnerable patterns and classify security weaknesses. The framework integrates continuous software testing practices to provide early vulnerability detection throughout the development lifecycle. Experimental evaluation demonstrates improvements in vulnerability detection accuracy, code analysis efficiency, false-positive reduction, and secure software testing capabilities. The proposed approach provides an intelligent solution for scalable software security management. Keywords— Graph Neural Networks, Machine Learning, Source Code Analysis, Vulnerability Detection, Secure Software Testing, Software Security, DevSecOps, Artificial Intelligence.