A Comprehensive Software Vulnerability Dataset Based on OWASP Top Ten Standard

Software vulnerabilities pose a critical threat to the security and reliability of modern systems, with the increasing sophistication of cyberattacks necessitating robust detection and mitigation mechanisms. This work presents a novel, publicly available dataset collected to study and enhance software vulnerability, aligned with the OWASP Top Ten security standard. We systematically collect and differentiate vulnerable and fixed code samples from GitHub commits across multiple programming languages by leveraging an automated web crawling pipeline. Our methodology integrates keyword mapping, graph-based representation using Code Property Graphs (CPGs), and automated parsing with Joern to provide textual and structural insights into code vulnerabilities. Unlike existing datasets, our resource includes one-to-one mappings between vulnerable and fixed code versions, addressing scale, diversity, and structure limitations. We evaluate the dataset using five machine learning models: Naive Bayes, Random Forest, Support Vector Machine, Gradient Boosting, and Multilayer Perceptron, which demonstrate its utility for both binary and multiclass classification tasks. Our results reveal that Gradient Boosting and Random Forest outperform other models, highlighting the dataset’s efficacy for real-world vulnerability study. This work establishes a foundation for further research into the automated collection of labeled vulnerability datasets and their application in improving cybersecurity tools and techniques.