An empirical study of the sensitivity of quality indicator for software module clustering

Abstract

Recently, there has been a significant progress in applying evolutionary multiobjective optimization techniques to solve software module clustering problem. The results of evolutionary multiobjective optimization techniques for software module clustering problem are a set of many non-dominating clustering solutions. Generally, the quality indicators of clustering solutions produced by these techniques are sensitive to minor variation in the decision variables of the clustering solutions. Researchers have focused on finding software module clustering with better cluster quality indicator; however in practice developers may not always be interested to better quality indicator clustering solutions, particularly if these quality indicators are quite sensitive. Under such situations, developer looks for clustering solutions whose quality indicators are not sensitive to small variations in the decision variables of the candidate clustering solution. The paper performs an experiment for the sensitivity analysis of quality indicator on software module clustering solution with two multiobjective formulations MCA and ECA. To perform the experiment the NSGA-II is used as multi-objective evolutionary algorithm. We evaluate sensitivity of quality indicators for six real-world software and one random problem. Results indicate that the quality indicator for MCA formulation is less sensitive than ECA formulation and hence MCA will be a better choice for multiobjective software module clustering from sensitivity perspective.