| Summary: | With the evolution of software systems, their size and complexity are rising rapidly. Identifying vulnerabilities as early as possible is crucial for ensuring high software quality and security. Just-in-time (JIT) vulnerability prediction, which aims to find vulnerabilities at the time of commit, has increasingly become a focus of attention. In our work, we present a comparative study to provide insights into the current state of JIT vulnerability prediction by examining three candidate models: CC2Vec, DeepJIT, and Code Change Tree. These unique approaches aptly represent the various techniques used in the field, allowing us to offer a thorough description of the current limitations and strengths of JIT vulnerability prediction. Our focus was on the predictive power of the models, their usability in terms of false positive (FP) rates, and the granularity of the source code analysis they are capable of handling. For training and evaluation, we used two recently published datasets containing vulnerability-inducing commits: ProjectKB and Defectors. Our results highlight the trade-offs between predictive accuracy and operational flexibility and also provide guidance on the use of ML-based automation for developers, especially considering false positive rates in commit-based vulnerability prediction. These findings can serve as crucial insights for future research and practical applications in software security.
|
|---|