Research Of Hybrid Clustering Ensemble Approaches

Ensemble learning, which has the ability to combine multiple classifiers or clustering learnersin order to notably improve the generalization capability and accuracy of the learning system,is the research focus in machine learning. Clustering ensemble, as an indispensable part ofensemble learning, has been applied to many fields. There are many proposed clusteringmethods. They are capable of handling the clustering problems based on low-dimensionaldata sets, while often lose the performance confronting high-dimensional, noisy, or largesample problems in practical situations. This is due to the weaknesses of traditional clusteringensemble approaches and difficulties in these practical problems:1)in high-dimensionalclustering ensemble process, plenty of trivial attributes erode the performance of the learnerand extend the running time;2)traditional clustering ensemble approaches are bad at handlingnoisy data sets;3)it is time-consuming to combine the clustering solutions for large sampledata sets. Aiming at covering the above shortcomings, this article proposed2workablemethods:1)clustering the attributes in order to reduce the active attributes and exclude thenoisy attributes;2)engaging structure ensemble strategy to simplify the combination ofclustering solutions in order to reduce the running time. In experiment section, we separatelyutilized synthetic data sets, public gene expression data sets, and UCI machine learning datasets to measure the performances of the proposed2methods. Results demonstrate theefficiency and effectiveness of these2methods.