Ensemble clustering via Fuzzy c-Means
Xin Wan, Hao Lin, Hong Li, Guannan Liu, Maobo An · 2017
Ensemble clustering is to fuse several basic partitions to find a single best cluster structure of data. With the prevalence of heterogeneous data rising from various application domains, ensemble clustering has become a state-of-the-art solution for cluster analysis due to its robustness and generalizability. However in the area of fuzzy systems, systematic research along this line is still in its initial stage. Finding a fuzzy consensus partition from multiple fuzzy basic partitions in an flexible and robust way remains a challenging yet promising issue. To this end, we propose Fuzzy Consensus Clustering (FCC), an ensemble clustering framework via Fuzzy c-Means, which fuses several fuzzy basic partitions from a utility perspective. Specifically, we first use the novel fuzzified contingency matrix to define the objective function of FCC. Then we derive a family of utility functions called FCCU that can transform FCC to a weighted piecewise fuzzy c-means clustering (piFCM) problem, which helps to establish an algorithmic framework for FCC with flexible choice of utility functions. Extensive experiments were conducted on various real-world data sets to validate the effectiveness of FCC. The results show that our method consistently outperforms baselines of traditional single clustering in terms of clustering quality.