Fuzzy c-Means Classifier for Relational Data

This paper proposes a relational version of the fuzzy c-means (FCM) classifier in which relational data instead of object data are used. The classifier based on the relational clustering is called "relational classifier". The classifier is useful when a feature space has an extremely high dimensionality that exceeds the number of objects and many of the feature values are missing, or when only relational data are available instead of the object data. The relational data is represented by a matrix in terms of distances (dissimilarity) between object data, and is not concerned with the relational database. The clustering algorithm used in the classifier includes, as a special case, the relational dual of FCM proposed by Hathaway, Davenport and Bezdek and can be seen as a simultaneous application of multidimensional scaling and clustering. The computational intensity of the classifier is comparable to Gaussian mixture classifier (GMC). The proposed classifier outperforms well established relational classifier known as k-nearest neighbor (k-NN) on several benchmark datasets from the UCI ML repository