Fuzzy density based clustering with generalized centroids

Fuzzy density-based clustering has been a challenge. Research has been focused on fuzzyfying the DBSCAN algorithm. Different methods have been proposed that use a fuzzy definition of core points within the DBSCAN algorithm. Our approach adapts the membership degree calculation known from fuzzy c-means by replacing the need for a distinguished centroid point by a more general cluster skeleton. These skeletons represent the clusters' shapes more accurately than a single point. We show how membership degrees can be calculated and that the resulting partitioning matrices allow the selection of more favorable parameters in the clustering process.