A Stability Analysis of Clustering Algorithms

The use of clustering for developing a description of a software system's architecture is fairly recent. Thus there is a need to evaluate various clustering algorithms and identify the ones which are expected to give good results for software. A criterion that can be used for evaluation is stability. A clustering algorithm is said to be stable if its output i.e. the clusters it produces, do not change drastically when small changes are made to the input data. The notion of stability is especially important when a software clustering algorithm is used to build a high level description of a software system that is currently under development. In this paper, we compare the stability of six hierarchical clustering algorithms by carrying out experiments on three open-source software systems. We also present an analysis of the results, which provides insight into the clustering process of the various algorithms