An Implementation of Just-in-Time Fault-Prone Prediction Technique Using Text Classifier

Since the fault prediction is an important technique to help allocating software maintenance effort, much research on fault prediction has been proposed so far. The goal of these studies is applying their prediction technique to actual software development. In this paper, we implemented a prototype fault-prone module prediction tool using a text-filtering based technique named "Fault-Prone Filtering". Our tool aims to show the result of fault prediction for each change (i.e., Commits) as a probability that a source code file to be faulty. The result is shown on a Web page and easy to track the histories of prediction. A case study performed on three open source projects shows that our tool could detect 90 percent of the actual fault modules (i.e., The recall of 0.9) with the accuracy of 0.67 and the precision of 0.63 on average.