Based On The Clustering Of The Program Software Defect Prediction Method Study

Software Quality Assurance (SQA) is one of the most important steps of software engineering, which evaluates the quality of software products by using a series of assurance methods to assure to deliver a perfect software product to the manager and customers. However, since the dual pressure of time to market and cost control, software development managers often find it hard to allocate sufficient time and resources for software test and other quality assurance activities. This always results in the defect-prone products to customers, which often leads to the error in the software running and even cause serious problems in system dependability and safety.There is some research work on the software defect prediction. Some is to build software defect prediction models by empirical study, which has been proved as a cost-effective way to alleviate this problem. With the empirical study on the metrics of software artifact or software development process and the defect data, researchers can build the defect prediction models to predict the defects number or the likelihood a software file or class contains defect. By using the defect prediction models, the defects contained in the software products can be predicted, which can assist the software developers in the software quality assurance and testing more effectively.The research of this paper focuses on the software defect prediction models, especially the models based on the software product metrics. The prior research in this direction mainly predicts defects by class or file, i.e. treat classes/files as basic units for defect evaluation and ranking, and improves the performance of the prediction models based on class/file. Some work proposes to predict defects on the higher package level and gets better recall and precision. However, it is also reported that package-based prediction models are less effective than class-based prediction models as far as the effort is concerned.In this paper, we propose a novel software defect prediction method based on functional clusters of programs. In the method, we use proper-grained and problem-oriented program clusters as the basic units of defect prediction in order to improve the performance, especially the effort-aware performance, of defect prediction.To evaluate the effectiveness of the method, we conducted an experimental study on Eclipse3.0system. We employed different data analysis methods to build the cluster-based defect prediction models and compared the performance with class-based defect prediction models. We found that, comparing with class-based models, cluster-based prediction models significantly improve the recall and precision of defect prediction. Moreover, according to the effort-aware evaluation, the cluster-based models can also improve the performance of defect prediction as far as the effort is concerned.