Home // FASSI 2023, The Ninth International Conference on Fundamentals and Advances in Software Systems Integration // View article

A Machine Learning-based Impact Analysis Tool and its Improvement Using Co-occurrence Relationships

Authors:
Teppei Kawabata
Tsuyoshi Nakajima
Shuichi Tokumoto
Ryota Tsukamoto
Kazuko Takahashi

Keywords: impact analysis; change requests; machine learning; co-occurrence relationships;

Abstract:
In the development of diverted software, impact analysis, which determines the extent of software impact on change requests, is an important task because it greatly affects the quality and efficiency of the development. We proposed a method that machine-learns the modification histories of the projects using word-embedding techniques and multi-label classifiers to accurately generate a ranking list of modification candidates of the software components in order of their sigmoid values. To improve accuracy of the method, this paper proposes to use the multi-label classifier algorism to take co-occurrence between labels into account because of the assumption of the dependencies between the components. Experiments were conducted on actual project data to compare the accuracy of the four algorisms: Convolutional neural networks, BR method, LP method, and RAkEL method. The result shows that RAkEL method, which takes co-occurrence relationships into account and does not over-learn, has the best accuracy among them.

Pages: 1 to 6

Copyright: Copyright (c) IARIA, 2023

Publication date: September 25, 2023

Published in: conference

ISSN: 2519-8475

ISBN: 978-1-68558-096-4

Location: Porto, Portugal

Dates: from September 25, 2023 to September 29, 2023