Can Faulty Modules Be Predicted by Warning Messages of Static Code Analyzer?

We have proposed a detection method of fault-prone modules based on the spam filtering technique, “Fault-prone filtering.” Fault-prone filtering is a method which uses the text classifier (spam filter) to classify source code modules in software. In this study, we propose an extension to use warning messages of a static code analyzer instead of raw source code. Since such warnings include useful information to detect faults, it is expected to improve the accuracy of fault-prone module prediction. From the result of experiment, it is found that warning messages of a static code analyzer are a good source of fault-prone filtering as the original source code. Moreover, it is discovered that it is more effective than the conventional method (that is, without static code analyzer) to raise the coverage rate of actual faulty modules. 1. Introduction Recently, machine learning approaches have been widely used for fault-proneness detection [1]. We have introduced a text feature-based approach to detect fault-prone modules [2]. In this approach, we extract text features from the frequency information of words in source code modules. In other words, we construct a large metrics set representing the frequency of words in source code modules. Once the text features are obtained, the Bayesian classifier is constructed from text features. In the fault-prone module detection of new modules, we also extract text features from source code modules, and Bayesian model classifies modules into either fault-prone or nonfault-prone. Since less effort or cost needed to collect text feature metrics than other software metrics, it may be applied to software development projects easily. On the other hand, since this approach accepts any input with text files, the accuracy of prediction could be improved by selecting appropriate input other than raw source code. We then try to find another input but source code. In this study, we use warning messages of a static code analyzer. Among many static code analyzers, we used PMD in this study. By replacing the input of fault-prone filtering from raw source code to warning messages of PMD, we can get the results of prediction by PMD and fault-prone filtering. The rest of this paper is organized as follows. Section 2 describes the objective of this research. Section 3 shows a brief summary of the fault-prone filtering technique with PMD. In Section 4, the experiments conducted in this study are described. Section 5 discusses the result of the experiments. Finally, Section 6 concludes this study. 2. Objective 2.1. Fault-Prone Module