An Attribute Selection For Severity Level Determination According To The Support Vector Machine Classification Result

Abstract

Determination of bug severity level is needed in fixing bug. Actually, in bug-tracking system, there is around 14 attributes used for defining a bug. But, all this time we do not know which attributes are highly influential for this. In this research, a new model of severity type classification using Infogain method for Bugzilla is proposed. As for the classsification process, we use Support Vector Machine, because this method is suitable in handling a massive data records. In this research, 8 bug attributes and 17.746 record of bug reports are involved. From the result of the experiment, we recommend five attributes which can be used effectively in classifying the severity types with a minimal value of infogain 0,33 which is component, qa_contact, summary, cc_list and product. The combination of those 5 attributes resulting in 99,83% accuracy of severity types classification. Keywords- Bug Tracking System; Severity Level Classification; TF-IDF; Infogain; SVM.