Research On Software Evolvability Prediction Based On Semi-supervised Data Grouping

Software evolution is an inevitable process in the software development lifecycle.Software maintainability prediction is a prerequisite for deciding whether or not to perform evolution activities.Accurate software maintainability prediction plays an important role in decision support and guidance in the evolution process.Software maintainability prediction is used to construct the prediction model for ascertaining the difficulty of software evolution.However,the current research on software maintainability prediction faces the common problem of imbalanced data distribution in the software maintainability dataset.Therefore,after more than 20 years of development,the analysis accuracy is always low,and since the training data marking process requires significant overhead,results from the existing method cannot be applied to the actual evolution activities.Aimed at solving the above problems,based on object-oriented(00)software system,we propose a semi-supervised software maintainability prediction method(SGMDH)in this paper.The SGMDH uses the cluster sampling method to solve the problem of imbalanced data distribution in the software maintainability dataset,and uses the partial labeled data samples to construct the prediction model by means of label propagation based semi-supervised learning and the group method of data handling algorithm.Meanwhile,we also construct the ensemble prediction models based on the ensemble learning.Based on the experiment on the user interface management system(UIMS)dataset and the quality evaluation system(QUES)dataset,the results show that,compared to the traditional non-sampled baseline method,the sample-based semi-supervised model and the integrated model all show better prediction performance.And the SGMDH model and the decision tree-based ensemble model presented in this paper achieves a more remarkable and excellent maintainability prediction ability.However,the semi-supervised model has stronger generalization ability and higher application value than the integrated model.Thus it verifies the effectiveness of the SGMDH method proposed by this paper in software maintainability prediction.