Porównanie stabilności taksonomii spektralnej oraz zagregowanych algorytmów taksonomicznych

High accuracy of the results is very important task in any grouping problem (clustering). It determines effectiveness of the decisions based on them. Therefore in the literature there are proposed methods and solutions that main aim is to give more accurate results than traditional clustering algorithms (e.g. k-means or hierarchical methods). Examples of such solutions can be cluster ensembles or spectral clustering algorithms. A desirable quality of any clustering algorithm is also stability of the method with respect to small perturbations of data (e.g. data subsampling, small variations in the feature values) or the parameters of the algorithm (e.g. random initialization). Empirical results shown that cluster ensembles are more stable than traditional clustering algorithms. Here, we carry out an experimental study to compare stability of spectral clustering and cluster ensembles.