A machine learning-based static analysis warning prioritization

Static analysis tools (SATs) can automatically detect software defects by analyzing the software source code. However, there are a large number of false positives in the warnings generated by SATs, which hinder the usability of SATs. To address this problem, this paper proposes an approach to constructing a machine learning-based warning prioritization model. Our approach explores the 91 warning features and obtains an optimal feature set with 25 features. Subsequently, the machine learning model is applied to learn a warning prioritization model based on the optimal feature set. To evaluate the warning prioritization model, we conduct the experimental evaluation on a real, open source, and large-scale warning dataset. The experimental results show that our approach is better than the random ranking algorithm. On average, in comparison with the random ranking algorithm, our approach reduces the number of false positives by 30.56% – 48.18% when developers check the warning list to find each true defect.