Software Defect Reports Severity Prediction

Severity analysis of software defect reports plays a key role in software maintenance.With the increasing scale of software,more and more software companies use bug tracking systems such as Bugzilla and Jira to manage software defects.These systems record software defects with defect reports,and realize reasonable allocation of software maintenance resources and reduce software maintenance costs by analyzing the severity of defect reports.Combined with the severity attribute of the defect report,existing studies mainly based on the text information of the defect report and completed the prediction of the severity of the defect report through the classification algorithm.However,the current research mainly focuses on the coarse-grained binary classification and lacks empirical analysis and fine-grained prediction of the severity of defects in actual software projects.To help developers to better analyze the seriousness of the defect report,based on the gravity two aspects of the empirical analysis and fine-grained predictions,the defect report serious analysis of the core problem:to solve the problem of severity empirical analysis,the analysis framework,the influence of gravity defect report proposed to evaluate the external influence of severity attribute;In order to solve the problem of severity prediction,a domain-feature-oriented coarse-grained severity prediction method and a fine-grained severity prediction method based on feature sequence reconstruction are proposed to improve the effectiveness of severity prediction of defects in different scenarios.First,this paper presents a framework for analyzing the severity impact of defect reporting.By analyzing the data of typical open source projects(Mozilla and Eclipse),as well as the distribution of the severity of defects in different components and projects,several interesting phenomena were found.Defects with normal severity take the shortest time to solve;The higher the number of defects the developer holds,the lower the repair rate.Secondly,this paper proposes a new domain feature-oriented coarse-grained severity prediction method.This part of the work will be divided into seven kinds of severity of software defect report seriously and not serious,two classes in the four kinds of feature selection algorithm and the result of 4 kinds of machine learning algorithm to deal with cross contrast,on the basis of the proposed new severity binary prediction method,the method USES T01 vectorization method dealing with the original text data sets,using the correlation coefficient of feature selection algorithm to optimize the characteristic of original data sets,using polynomial bayesian algorithm to training and testing data sets,make predictions AUROC value increased to 0.7642.Finally,a sorting-based feature sequence reconstruction method for fine-grained severity prediction is proposed.Because the binary severity classification(serious and minor)of defect reports cannot accurately express the severity of defect reports,in order to solve the fine-grained severity prediction problem,this paper USES the feature subset optimization idea of defect reports and proposes the feature subset optimzation method.This method reduces the subset of severity characteristics of software defect report by 50%,and at the same time,significantly improves the prediction effect of fine-grained defect report severity,and increases its weighted f-value index to 76.7%.This study not only makes up for the deficiency of empirical analysis on the severity of defect reports,but also further improves the performance of quantitative prediction of fine-grained severity,which is of certain reference value for rational allocation of resources in software maintenance process.