A novel approach to robust fuzzy clustering of relational data

Abstract

We propose a new approach to robust fuzzy clustering of relational data, which does not require any particular restriction on the relation matrix. More precisely, we adopt an algorithm based on the Fuzzy C-Means (FCM) algorithm, improved with Dave's concept of Noise Cluster, and suitable for data which are expressed in terms of mutual numerical relationships among patterns. In this way, we tackle a relational clustering problem taking advantage of the stability and effectiveness of object data clustering algorithms. We also exploit the concept of prototype as representative of the mutual relationships of a group of similar patterns. We show that our approach is more scalable and less sensitive to cluster initialization and parameter variations than the robust Non-Euclidean Fuzzy Relational data Clustering algorithm (robust-NE-FRC), one of the most efficient recently proposed relational algorithms, on both real and synthetic data sets.