Classificação automática da prioridade de defeitos utilizando seleção de atributos: um estudo de caso na indústria

Abstract

Defects are inevitable in software projects, so adopting a policy of analyzing and managing defects during the software development cycle is vital for quality assurance. Storing defect reports is commonplace in the software development cycle, but the information contained in the reports is difficult to understand because they are usually written in natural language. In this sense, classifying defects can help streamline the defect management process. As they are written in natural language, automatic defect classification becomes difficult and may have low effectiveness, with an analyst taking an average of 6 minutes for each bug report. Some studies propose using feature selection approaches to increase the accuracy of automatic defect classification. This study presents an industry case study with automatic defect classification using feature selection approaches proposed in the literature. In addition, a change to the USES feature selection algorithm is proposed to improve its effectiveness, the new algorithm was denominated USES+. As a result, we have that the automatic feature selection approaches result in a greater effectiveness of the automatic classification of defect written in natural language. Finally, USES+ obtained a better effectiveness when compared to USES, even with significant differences. Thus, in general, automatic defect classification using feature selection approaches has shown to be promising with good effectiveness. In this way, automatic classification allows reduction of the cost of prioritization of defects, significantly improving the delivery of the essential corrections.