What Strokes to Modify in the Painting? Code Changes Prediction for Object-Oriented Software

Abstract

Software systems shall evolve to fulfill users’ increasingly various and sophisticated needs. As they become larger and more complex, the corresponding testing and maintenance have become a practical research challenge. 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. A new software release may be based on several previous releases. Thus, 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. Based on a series of change evolution matrices, a code change prediction approach is proposed based on the Gauss Process Regression (GPR) algorithm. Experiments are conducted on 17 software systems collected from GitHub to evaluate our prediction approach. The results show that our approach outperforms three existing state-of-the-art approaches with significantly higher prediction accuracy.