Research On Svm-based Evaluation Method Of Clone Code Harmfulness

Clone code (also known as duplicated code) has always been a popular researchfield of software engineering. With the limitations of attention on single versionsoftware, traditional view is that clone code is harmful to the program and should bepromptly detected and refactored. However, a few studies in recent years found thatcloned code is not necessarily harmful-in the process of code evolution, some clonecodes exist for a very short life cycle, while others which never changed in the systemhas strong robustness, and it is unwise to refactor them blindly. Therefore, consideringthe stability of the system, maintenance costs, refactoring difficulty and other factors, itis very necessary to comprehensively evaluate the harmfulness of clone code.Unfortunately, there are very few such studies so far.To solve the problem of clone code harmfulness evaluation, we propose a SVMbased method with both static metrics and evolution metrics to evaluation theharmfulness of clone code. Referring to the research method of software defectprediction, our method regards the evaluation of the clone code harmfulness as asupervised classification problem in machine learning. First, we proposed a standarddefinition of clone code harmfulness and corresponding sample labeling method.Second, referring to correlation research on software defects prediction and clone codeevolution, we proposed two software metrics-static metrics and evolution metrics tocharacterize the information of clone code, and chose the SVM as the core algorithm ofour evaluation model. Finally, with this model we train and test on the clone codesamples with both static metrics and evolution metrics. After cross-validation andparameter optimization, the work of our clone code harmfulness evaluation model hascompleted.In the last section of this paper, we experiment on6different types of open-sourcesoftware system written in3kinds of programming languages to verify and assess ourmethod.The experiment results show that the proposed method of this paper has betterapplicability and higher accuracy. It is a meaningful attempt to evaluate the harmfulnessof clone code. In addition, the analysis on the influence of the proposed metrics alongwith our experiment results also provides a valuable reference for the future study.