Predicting change propagation using domain-based coupling

Aryani, A 2013, Predicting change propagation using domain-based coupling, Doctor of Philosophy (PhD), Computer Science and Information Technology, RMIT University.

Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Aryani.pdf Thesis Click to show the corresponding preview/stream application/pdf;... 1.90MB
Title Predicting change propagation using domain-based coupling
Author(s) Aryani, A
Year 2013
Abstract Most enterprise systems operate in domains where business rules and requirements frequently change. Managing the cost and impact of these changes has been a known challenge, and the software maintenance community has been tackling it for more than two decades. The traditional approach to impact analysis is by tracing dependencies in the design documents and the source code. More recently the software maintenance history has been exploited for impact analysis.

The problem is that these approaches are difficult to implement for hybrid systems that consist of heterogeneous components. In today’s computer era, it is common to find systems of systems where each system was developed in a different language. In such environments, it is a challenge to estimate the change propagation between components that are developed in different languages. There is often no direct code dependency between these components, and they are maintained in different development environments by different developers. In addition, it is the domain experts and consultants who raise the most of the enhancement requests; however, using the existing change impact analysis methods, they cannot evaluate the impact and cost of the proposed changes without the support of the developers.

This thesis seeks to address these problems by proposing a new approach to change impact analysis based on software domain-level information. This approach is based on the assumption that domain-level relationships are reflected in the software source code, and one can predict software dependencies and change propagation by exploiting software domain-level information. The proposed approach is independent of the software implementation, inexpensive to implement, and usable by domain experts with no requirement to access and analyse the source code.

This thesis introduces domain-based coupling as a novel measure of the semantic similarity between software user interface components. The hypothesis is that the domain-based coupling between software components is correlated with the likelihood of the existence of dependencies and change propagation between these components. This hypothesis has been evaluated with two case studies:

• A study of one of the largest open source enterprise systems demonstrates that architectural dependencies can be identified with an accuracy of more than 70% solely based on the domain-based coupling.

• A study of 12 years’ maintenance history of the five subsystems of a significant sized proprietary enterprise system demonstrates that the co-change coupling derived from over 75,000 change records can be predicted solely using domain-based coupling, with average recall and precision of more than 60%, which is of comparable quality to other state-of-the-art change impact analysis methods.

The results of these studies support our hypothesis that software dependencies and change propagation can be predicted solely from software domain-level information. Although the accuracy of such predictions are not sufficiently strong to completely replace the traditional dependency analysis methods; nevertheless, the presented results suggest that the domain-based coupling might be used as a complementary method or where analysis of dependencies in the code and documents is not a viable option.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Computer Science and Information Technology
Keyword(s) Software Maintenance
Software Evolution
Change Impact Analysis
Domain-Based Coupling
Change Propagation
Version Filter Type
Access Statistics: 344 Abstract Views, 480 File Downloads  -  Detailed Statistics
Created: Tue, 18 Jun 2013, 08:58:12 EST by Brett Fenton
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us