An intuitionistic fuzzy representation based software bug severity prediction approach for imbalanced severity classes

Abstract

In order to improve software reliability and quality, the triager must assess the severity of the software bug and allocate suitable resources on time. However, the triager faces many difficulties in understanding various software bugs that involve lots of uncertainty and irregularities. Additionally, it can be challenging for the triager to determine the severity of bugs that are semantically close to multiple severity labels. To address these problems, a topic modeling and intuitionistic fuzzy similarity measure-based software bug severity prediction technique (IFSBSP) is proposed in this paper. Initially, the Synthetic Minority Oversampling Technique (SMOTE) is applied to balance the severity classes in software bug repositories. Then topic modeling is used to generate topics based on the probability of underlying uncertainty in software bugs. Using these topics, the intuitionistic fuzzy membership, non-membership, and hesitancy membership degrees of a software bug are calculated for multiple severity labels. Then, 15 IFS techniques are investigated for a new bug in order to compute its similarity to multiple severity labels. The Eclipse, Mozilla, Apache, and NetBeans software bug repositories are used to evaluate the performance of IFSBSP and the state-of-the-art models. On these software bug repositories, the IFSBSP model outperforms state-of-the-art models by achieving accuracy of 91.6%, 90.9%, 88.1%, and 92.9% and an F-measure of 90.7%, 91.1%, 89.3%, and 91.7%, respectively.