Clustering Data Retrieved from Java Source Code to Support Software Maintenance: A Case Study

Data mining is a technology recently used in support of software maintenance in various contexts. Our works focuses on achieving a high level understanding of Java systems without prior familiarity with these. Our thesis is that system structure and interrelationships, as well as similarities among program components can be derived by applying cluster analysis on data extracted from source code. This paper proposes a methodology suitable for Java code analysis. It comprises of a Java code analyser which examines programs and constructs tables representing code syntax, and a clustering engine which operates on such tables and identifies relationships among code elements. We evaluate the methodology on a medium sized system, present initial results and discuss directions for further work.