The Research On Blocking Bug Prediction Approach Based On XGBoost Algorithm

With the increasing number of software projects,software quality has become increasingly crucial.Therefore,many researchers and engineers in the field of software engineering are focusing on software bug management tasks,only few researchers pay attention to the blocking bug prediction problem.The modules repaired by developers depend on other modules which have blocking bugs,and blocking bugs need more time to be repaired.Therefore,automatically predicting whether a software bug is a blocking bug will allow developers or testers to predict such bugs in advance and take corresponding countermeasures(such as reducing the degree of coupling of such bugs),which can help reduce the negative impact of blocking bugs on software development and fixing.Unfortunately,current studies have extracted corresponding features from the bug reports to represent the bug reports,but they did not consider to extract more features to better represent them.In addition,current studies utilized supervised algorithms to implement this task,and did not consider the dependencies among individual classifiers,thus the blocking bugs cannot be predicted more accurately.In order to resolve these problems,this paper proposes a method of blocking bug prediction named XGBlocker based on the enhanced dataset and XGBoost algorithm to automatically predict blocking bugs.The main work and contributions are as follows:14 features were first collected from the pre-defined fields,description and comments of the bug report.Since fewer features cannot better characterize a bug report,this method adds new features and combines them with the original 14 features to characterize a bug report,thereby establishing an enhanced data set.This can capture more information in the sample and thus better distinguish between different samples.Aiming at the dependency problem between individual classifiers,a blocking bug prediction model based on XGBoost algorithm is proposed.There is a correlation among individual classifiers of this model.In this paper,all the features in the first stage are input into the model to perform the task of blocking bug prediction.This paper conducts experimental verification research on XGBlocker.In order to evaluate the effectiveness of this method,this paper collects bug reports from four open source projects.Based on these data,this paper performs performance evaluation and analysis of the method.Experimental results show that the method proposed in this paper outperforms the baseline method in three performance indicators(F1-score,ER@20% and AUC),which verifies that the method proposed in this paper,XGBlocker,can predict blocking bugs more accurately,and can help developers reduce workload and save working time.