Fuzzy clustering in parallel universes

We propose a modified fuzzy c-means algorithm that operates on different feature spaces, so-called parallel universes, simultaneously. The method assigns membership values of patterns to different universes, which are then adopted throughout the training. This leads to better clustering results since patterns not contributing to clustering in a universe are (completely or partially) ignored. The outcome of the algorithm are clusters distributed over different parallel universes, each modeling a particular, potentially overlapping, subset of the data. One potential target application of the proposed method is biological data analysis where different descriptors for molecules are available but none of them by itself shows global satisfactory prediction results. In this paper we show how the fuzzy c-means algorithm can be extended to operate in parallel universes and illustrate the usefulness of this method using results on artificial data sets.