Research On The Fuzzy Spectral Partition Clustering Ensemble Algorithms Based On Occasion

As an important basic technology method of data mining and machine learning, clustering analysis is widely applied in pattern recognition, image segmentation, social computing and other fields. Among, the scholars at home and abroad research on fuzzy clustering algorithms including the fuzzy C-means clustering algorithm from different angles, and put forward a series of new fuzzy clustering algorithms.Although clustering algorithms are various, the data in the real world also have a variety of structure and shape, so any clustering algorithm cannot accurately reveal the structure of the complex data. Clustering ensemble is in due time, combining multiple cluster members’results to get uniform and reasonable clustering results. In recent years, many studies have shown that clustering ensemble technology may effectively improve the single clustering algorithm’s accuracy and stability, one of the key problems is how to combine cluster members’results to get better clustering result.In this thesis, the concept of clustering ensemble occasion is proposed, and applying the early stopping rules to the clustering ensemble, puts forward the generation ways of clustering members based on the early stopping rules.What’s more, this thesis severally integrates in three stages results of spectral partition clustering ensemble algorithm,which are called partition integration, network integration and utility integration. And then, applying the FCM clustering algorithm to the following clustering phase of the above clustering ensemble algorithms, presents a set of fuzzy spectral partition clustering ensemble algorithms based on occasion.Through the theoretical analysis of the clustering ensemble algorithm based on the early stopping rules and the fuzzy spectral partition clustering ensemble algorithms based on occasion, and the corresponding experimental analysis, it may get some conclusions. One, the performance of clustering ensemble based on the early stopping rules is superior to that based on the end solutions of clustering members, while the former takes less time. The other, the fuzzy spectral partitioning clustering ensemble algorithms based on occasion in this thesis, compared with some graph partitioning clustering ensemble algorithms in existence, do well in the clustering effectiveness and the execution time. Especially, the fuzzy spectral partition clustering ensemble algorithm based on partition integration, not only effectively removes of ’noise’ and more accurately expresses of data sets structure, but also has particularly prominent performance in the execution time.