Assessing the Complexity of Software Bug Severity using Evolutionary SOM Clustering Approach

Abstract

As software demands increase and software delivery span decreased, ensuring software quality becomes a challenge. However, due to software complexity and inadequate testing, no software can claim to be error-free. Maintenance activities are therefore required to ensure that the software works smoothly. Software bug repositories are of great source of knowledge. When a bug is identified by a tester or software engineer, its related information is entered in Bug Tracking System. During its life time, a bug enters into various bug states for resolution. These bug tracking systems enable users to report the bugs found while the software is running. However, the severity prediction has recently gained a lot of attention in software maintenance. Bugs with higher severity should be corrected before bugs with lower severity. In this paper, an evolutionary interactive approach is proposed to analyze the bug reports
and assesses the severity. This paper presents a SBCC – Software Bug Complexity Cluster using Self Organizing Maps (SOM) approach. In this SBCC a feature matrix is constructed using bug durations and software bug complexities are grouped into different clusters including Blocker, Critical, major, trivial and minor by specifying severity using two different methods, namely- k-means and SOM. The accuracy of different bug complexities are estimated using bug duration, proximity error and different distance functions. Our systematic study has ascertained that the proximity error and fitness on SOM has better accuracy and performance compared to K-Means. The collected results showed that the SBCC attained better performance with respect to fitness and cluster proximity error.