Actionable Static Analysis Code Warnings Identification with SMOTE-based Classification

Automatic static code analysis warnings are produced by tools for static code analysis (like FindBugs), but they suffer from an excessive number of false alarms that are ignored by developers. Accurately identifying actionable static analysis code warnings helps developers be more efficient, and the static analysis tool can be made more useful. In reality, static analysis warning datasets have a class imbalance in actionable warning identification since there are a greater number of un-actionable (false alarms) than actionable alarms. Data sampling techniques address the issue of class imbalance in classification, which is a problem with the quality of the collected data. By producing synthetic instances that may be acted upon, the Synthetic Minority Oversampling Technique (SMOTE) and its variations assist in solving the problem of class imbalance. SVM, Random Forest, Naïve Bayes, and Decision Tree classifiers were created as classification models using the eight Java open-source software projects datasets (Cassandra, Jmeter, Commons.lang, Luence-solr, Ant, Tomcat, Maven, and Derby). The performance of the developed models is evaluated using precision, recall, f-measure, ROC and false positive rate. The analysis of the experimental findings revealed that the SMOTE not only addressed the issue of class imbalance but also enhanced the classification performance and reduced the false positive rate of the experimental models when it came to the identification of warnings.