Finding Building Blocks for Software Clustering

It is generally believed that good modularization of software leads to systems which are easier to design, develop, test, maintain and evolve [1]. Software clustering using search–based techniques has been well studied using a hill climbing approach [2,4,5,6]. Hill climbing suffers from the problem of local optima, so some improvement may be expected by by considering more sophisticated search-based techniques. However, hitherto, the use of other techniques to overcome this problem such as Genetic Algorithms (GA) [3] and Simulated Annealing [7] have been disappointing. This poster paper looks at the possibility of using results from multiple hill climbs to form a basis for subsequent search. The findings will be presented in the poster created for the poster session in GECCO 2003.