An empirical study on classification methods for alarms from a bug-finding static C analyzer

A classification method of approximating dynamic program of alarm states occurring at each program point and finding bugs, by examining the approximate states, from an automatic bug-finding static C analyzer, was analyzed. The classifier based on features extracted from the bug reports and which statically detected buffer-overrun errors in C programs, was attached to a C analyzer, Airac. The symptom results provided by Airac for alarms are syntactic symptoms, semantics symptoms, and result symptoms. The Receiver Operating Characteristic (ROC) curve of the classification methods show an the most effective classification methods for alarms are boosting, random forest, and Support Vector Machine (SVM) methods. The results also show that the trained classifiers for a range of 39.8% to 69.5% and with multiple codebases, are unbiased for Linux or non-Linux alarms.