An Integrated Approach to Detect Fault-Prone Modules Using Complexity and Text Feature Metrics

Early detection of fault-prone products is necessary to assure the quality of software product. Therefore, fault-prone module detection is one of the major and traditional area of software engineering. Although there are many approaches to detect fault-prone modules, they have their own pros and cons. Consequently, it is recommended to use appropriate approach on the various situations. This paper tries to show an integrated approach using two different fault-prone module detection approaches. To do so, we prepare two approaches of fault-prone module detection: a text feature metrics based approach using naive Bayes classifier and a complexity metrics based approach using logistic regression. The former one is proposed by us and the latter one is widely used approach. For the data for application, we used data obtained from Eclipse, which is publicly available. From the result of pre-experiment, we find that each approach has the pros and cons. That is, the text feature based approach has high recall, and complexity metrics based approach has high precision. In order to use their merits effectively, we proposed an integrated approach to apply these two approaches for fault-prone module detection. The result of experiment shows that the proposed approach shows better accuracy than each approach.