We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Software Engineering

Title: Assessing Validity of Static Analysis Warnings using Ensemble Learning

Abstract: Static Analysis (SA) tools are used to identify potential weaknesses in code and fix them in advance, while the code is being developed. In legacy codebases with high complexity, these rules-based static analysis tools generally report a lot of false warnings along with the actual ones. Though the SA tools uncover many hidden bugs, they are lost in the volume of fake warnings reported. The developers expend large hours of time and effort in identifying the true warnings. Other than impacting the developer productivity, true bugs are also missed out due to this challenge. To address this problem, we propose a Machine Learning (ML)-based learning process that uses source codes, historic commit data, and classifier-ensembles to prioritize the True warnings from the given list of warnings. This tool is integrated into the development workflow to filter out the false warnings and prioritize actual bugs. We evaluated our approach on the networking C codes, from a large data pool of static analysis warnings reported by the tools. Time-to-time these warnings are addressed by the developers, labelling them as authentic bugs or fake alerts. The ML model is trained with full supervision over the code features. Our results confirm that applying deep learning over the traditional static analysis reports is an assuring approach for drastically reducing the false positive rates.
Subjects: Software Engineering (cs.SE); Machine Learning (cs.LG)
Cite as: arXiv:2104.11593 [cs.SE]
  (or arXiv:2104.11593v1 [cs.SE] for this version)

Submission history

From: Hariharan Manikandan [view email]
[v1] Wed, 21 Apr 2021 19:39:20 GMT (244kb)

Link back to: arXiv, form interface, contact.