Using Developer Factors and Horizontal Partitioning to Recommend Bug Severity in Open-Source Software Projects

Software testing is performed during engineering of software to uncover the faults hidden in software systems. This data about the faults and failures is recorded in problem report forms. In addition to recording the symptoms of the problems, the problem report forms record severity of the fault. This severity of the potential failure refers to the impact on user or business. Usually, this severity is assigned by humans responsible for testing of software. Given the fact that faults in the software are result of human errors, literature has reported the use of developers' information to predict faults in software. Using the same intuition, this paper investigates the use of developers' information (along with the static code metrics) to determine the extent of high severity faults in software. To this end, three open-source software projects are studied in this paper and the code files are identified that have a chance of high severity faults. This recommendation can help testing teams identify fault prone files and label code files with high, medium, or low chance of severe faults. Static code metrics for the three projects are collected through SonarQube and developer related factors are extracted from the GitHub through custom scripts. This data is used to develop machine learning classifiers and code files are labelled as high, medium, and low number of severe faults. In order to further study the relationship between the number of developers and the files with severe faults, association between the developer factors and the presence of high severity faults is found. Study of this association reveals that if there are more developers working on a file, the file is more probable to have critical faults.