Research On Code Changes Prediction For Fast Iterative Applications

Software systems shall evolve to fulfill users' increasingly various and sophisticated environment.Especially in the context of large-scale,networked and fast iteration of software systems,the research on the evolution of applications and the prediction of source code changes has become a hot topic in the field of software engineering.In this paper,we employ an approach that can identify the change-proneness in the source code of new object-oriented software releases and predict the corresponding change sizes.We first define two metrics,namely Class Change Metric and Change Size Metric,to describe the features and sizes of code changes on the basis of comparative analysis and empirical study of existing metrics.Then,based on CCM and CSM,the source code changes of every two consecutive releases are mined,and a series of change evolution matrices containing information about class changes are established.We employ an Entropy Weight Method to calculate the best window size for determining the number of previous releases to use in the prediction of change-proneness in the new release.Next,a code change prediction approach is proposed based on the Gauss Process Regression(GPR)and Long Short-Term Memory(LSTM)algorithm.Experiments are conducted on 17 software systems and all versions of software systems collected from GitHub to evaluate our prediction approach.The results show that our approach outperforms PSM,QHC and the model based on C&K metric.