Partially Horizontal Collaborative Fuzzy C-Means

Abstract

Horizontal collaborative clustering is such a clustering method that carries clustering on one data set describing a pattern set in one feature space with collaborative introducing of outer partition information obtained by clustering on another data set but describing the same pattern set in another feature space. In order to implement the collaborative clustering, horizontal collaborative fuzzy c-means (HC-FCM) was proposed by W. Pedrycz. In HC-FCM, the outer partition matrix is incorporated with the objective function in FCM. This manner of making use of the outer partition matrix emphasizes on the use of total collaborative clustering information provided by the outer partition matrix, thus this method can be called completely horizontal collaborative fuzzy c-means (CHC-FCM). In reality, on many occasions of collaborative clustering, we may be interested only in the cluster information of some special patterns, say the patterns with distinct membership for example. In this paper, we implement the horizontal collaborative clustering with the partial supervision clustering approach where the clustering is carried by the guidance of some labeled patterns. In this approach, we can select the patterns we are interested in to provide FCM with collaborative information and control the degree of the influence of the selected patterns on the clustering. This new method is called partially horizontal collaborative fuzzy c-means (PHC-FCM). After presenting two approaches to realizing the selection of the labeled patterns, named cut-set based approach and entropy based approach, we give the detailed algorithm of PHC-FCM. Experiments are carried and show the performance of the new method.