Evaluating the suitability of data clustering for software remodularisation

The paper describes the investigation of a technique for remodularising legacy software; that of cluster analysis. This technique takes into account data cohesion as an influencing factor of the remodularisation process and compares and contrasts this with calling structure analysis. Cluster analysis is a well-established discipline used in other sciences but only recently linked to software remodularisation. A number of different cluster analysis techniques were chosen for evaluation. A tool was developed to perform this cluster analysis with two main aims; to provide a way of evaluating the chosen techniques and to provide a usable method of generating a remodularisation of a software system. The techniques evaluated produced modularisations of varying quality. However, it is thought that cluster analysis is a valuable and useful approach to software remodularisation that is worth further investigation. In particular the data structure analysis provided significantly better results than the calling structure analysis