How Starting Points and Representations Affect Software Modularisation: An Empirical Analysis

Exploring the software system via different clustering representations and starting points is not an easy task. Calculating the search space’s fitness with the existence of various representations and starting points would be challenging since every representation’s results vary due to several factors. This paper aims to exploit different automated software modularisation clustering representations by deploying Random Mutation Hill Climbing (search-based) algorithm and five starting points. Fitness functions, including EVM and EVMD, were tested on fifty open-source datasets ranging in different sizes and run for a million iterations to find the best software clustering combination of representation and starting point. The paper presents empirical results that compare the robustness of representations that led to yield some interesting insights related to the List of Lists.