PREDICTING BUG REPORTS WITH A NATURE-INSPIRED ENSEMBLE MACHINE LEARNING APPROACH

Abstract

Researchers have focused on the
maintenance process of software systems in
software development systems because of its
significance in resolving defects found
during software testing through the use of
bug reports (BRs), which contain
comprehensive information such as the bug's
description, status, reporter, assignee,
priority, and severity, among other details.
The primary challenge in this approach is
figuring out how to examine these BRs in
order to find every flaw in the system. Since
there are a lot more BRs, doing this by hand
is laborious and time-consuming. The
automated solution is thus the most
effective. The majority of current research is
on automating this process from various
angles, such determining the bug's
importance or severity. They did not,
however, take into account the fact that the
defect is a multi-class categorisation issue.
By putting out a novel prediction model to
examine BRs and forecast the kind of defect,
this study resolves this issue. Using machine
learning and natural language processing
(NLP) methods, the suggested model builds
an ensemble machine learning algorithm. A
publicly accessible dataset for two online
software bug repositories (Eclipse and
Mozilla) including six classes—Program
Anomaly, GUI, Network or Security,
Configuration, Performance, and TestCode—is used to replicate the suggested
approach. According to the simulation
findings, the suggested model can
outperform the majority of current models in
terms of accuracy, achieving 90.42%
without text augmentation and 96.72% with
text augmentation