New Strategies for Solving Cluster Ensemble Problems

One method of combining the results of multiple clusterings is the use of sequential cluster ensemble (SCE) algorithm. Although the result of thismethod is comparable to the other cluster combinationmethods, it has two major drawbacks. First its output is dependent to the order of input clusterings. Second theoutput is not robust to the existence of weak clusterings. To solve this problem we consider 2 new strategies. In the first strategy we proposed that by taking a hierarchical approach, it may be possible to improve the ensemble accuracy. We introduced a concept called‘N clusterings combination’ as a second strategy. The ‘N clusterings combination’ will be used instead of ‘2 clusterings combination’ of traditional SCE consensusmethod. With comprehensive experiments on several publicly available real datasets and Monte Carlo simulations, itis empirically demonstrated that using these strategiesthe quality of resultant clustering is usually better than the traditional SCE algorithm

کلیدواژه ها:

Clustering ، ensemble clustering ، clustering combination ، ensemble learning