Software modularization using the modified firefly algorithm

Software modularization is a significant way to increase the maintenance of software systems. Designers depict the structure of software systems as a directed graph referred to as Module Dependency Graph in which nodes correspond to the software modules (such as files and classes) and edges relate to their relations (like function calls and inheritance relationship); These graphs can be partitioned into the meaningful clusters to be more accessible. Since Graph Partitioning is considered as a NP problem, here the application of exhaustive search algorithms are not so cost-effective that even for very small graphs they are not applicable. Therefore, many random approaches have been proposed for solving the problem. In this paper, a Firefly based method is presented to partition the software systems. However, some adjustments are needed to be made for using the FA for the problem. These modifications are as follows: change of the definition of fireflies, how to move fireflies toward better positions, and the alteration of fireflies´ distance definition. Experimental results reveal that in most of the cases the proposed approach has a remarkable superiority over both the Genetic and the Learning Automaton algorithms in solving the problem.