Automated software bug severity classification using ensemble machine learning scheme: A real case study

Software bug report classification is one of the most significant processes in software development for determining the nature and severity of faults based on their causes and effects. In many projects, software experts implement this process manually, which requires exorbitant time and effort. Although there are a few studies on automatic bug report classification using machine learning techniques, they mainly focus on structured open-source datasets. This paper presents an ensemble learning approach utilizing various multiclass machine learning, text classification, and natural language processing techniques for automated software bug severity classification, with an application in the Persian language. This language, due to its unique characteristics, requires the adoption of different approaches from those applicable to the English language for text classification. The proposed approach utilizes a real bug dataset extracted from a case study containing unstructured bug reports. This dataset contains 4429 bug reports about the software product of the studied company, which is used by thousands of users in government and private organizations. These bug reports were recorded in Persian text by the testing team or software users, and then classified based on their severity through meetings of development team managers in the company. Results demonstrate that the developed appraoch is highly accurate and significantly faster than manual classification, which can dramatically decrease software development time and cost.