Modified affinity propagation clustering

Abstract

Affinity propagation clustering is an efficient clustering technique that does not require prior knowledge of the number of clusters. However, it sets the input preferences without considering data set distribution and competition in the former iteration is ignored when updating messages passing between data points. This paper presents a modified affinity propagation algorithm. Firstly, preference for each data point to serve as an exemplar is computed self-adaptively based on data set distribution; then encouragement and chastisement mechanism is introduced for updating message of availability. Experimental results on standard data sets and synthetic data sets demonstrate feasibility and effectiveness of the proposed algorithm.