Software fault prediction based on change metrics using hybrid algorithms: An empirical study

Abstract

Quality of the developed software depends on its bug free operation. Although bugs can be introduced in any phase of the software development life-cycle but their identification in earlier phase can lead to reduce the allocation cost of testing and maintenance resources. Software defect prediction studies advocates the use of defect prediction models for identification of bugs prior to the release of the software. Use of bug prediction models can help to reduce the cost and efforts required to develop software. Defect prediction models use the historical data obtained from software projects for training the models and test the model on future release of the software. In the present work, software change metrics have been used for defect prediction. Performances of good machine learning and hybrid algorithms are accessed in prediction of defect with the change metrics. Android project has been used for experimental purpose. Git repository has been used to extract the v4–v5, v2–v5 of android for defect prediction. Obtained results showed that GFS-logitboost-c has best defect prediction capability.

Keywords

• Testing
• Bug
• Defects
• Machine learning
• Hybrid algorithms
• Software metrics