An Effective Model to Predict the Extension of Code Changes in Bug Fixing Process Using Text Classifiers

After the issue tracking system (ITS) receives a bug report, this report is analyzed and triaged. To automate bug handling, past researches focused on topics such as automatic detection of duplication, prediction of fixing time, and automatic fault localization, to prioritize tasks, manage resources better and allocate resources more efficiently. However, predicting the amount and type of the changes necessary to fix a reported bug in the code, as soon as the bug report is received and before it is assigned to a programmer, has been neglected in previous researches. It seems that this prediction can be applied in different fields like bug fixing time, levels of integration test, etc. In this work, a model for predicting the amount of software code changes in the bug fixing process is presented. To achieve this, the problem is modeled as a text classification problem and solved using ensemble classifiers. The applicability of the proposed model is justified by statistical analysis. Our study shows that there is a significant relationship between the amount of changes in the software code and bug fix time. An empirical study by analyzing code changes in 15 projects conducted and tested with bagging-based and voting-based classifiers with different basic classifiers. The accuracy of the best model was measured approximately 72% on average.