Clustering is a data analysis method that creates groups of objects according to a similarity notion. Amongst the existing clus-tering algorithms, the possibilistic fuzzy c-means (PFCM) is a well-known algorithm since it generates a possibilistic partition. Such possibilistic partition is helpful in the presence of a noisy environment and allows to express various types of uncertainty and imprecision. In recent years, the performance of clustering methods has been improved by incorporating partial information. The approach, called semi-supervised clustering, introduces instance-level information such as labeled patterns in the clustering process. In this work, we propose to extend PFCM to combine labeled patterns with the possibilistic framework. To provide more flexibility to the new method, in addition to the Euclidean distance, an adaptive distance measure is considered. Experimental results show the interest of our new semi-supervised possibilistic fuzzy c-means algorithm on various data sets. (c) 2022 Elsevier B.V. All rights reserved.
展开▼
