Convex fuzzy k-medoids clustering
Fuzzy clustering, Convex optimization, Multiple representatives.
The k-medoids model is one of the most popular clustering methods. In this work, we propose the Convex Fuzzy k-Medoids Problem (CFKM), which not only allows one object to be assigned to multiple clusters, but also allows a cluster to be represented by multiple medoids. The proposed model is convex and thus is robust to initialization. To evaluate the importance of CFKM, we compare it with another two fuzzy k-medoids models: the Fuzzy k-Medoids Problem (FKM) and the Fuzzy clustering with Multi-Medoids Problem (FMMdd), both solved by heuristics due to their computational complexity. Experiments with both synthetic and real-world data, along with an user survey, show that CFKM is not only more robust to the choice of parameters of fuzzy models, but also is the only able to reveal important aspects of inherently fuzzy data.