Affinity propagation: An exemplar‐based tool for clustering in psychological research

Abstract

Affinity propagation is a message-passing-based clustering procedure that has received widespread attention in domains such as biological science, physics, and computer science. However, its implementation in psychology and related areas of social science is comparatively scant. In this paper, we describe the basic principles of affinity propagation, its relationship to other clustering problems, and the types of data for which it can be used for cluster analysis. More importantly, we identify the strengths and weaknesses of affinity propagation as a clustering tool in general and highlight potential opportunities for its use in psychological research. Numerical examples are provided to illustrate the method.

Unsupervised acoustic classification of individual gibbon females and the implications for passive acoustic monitoring

“…Unlike many other commonly used unsupervised clustering algorithms, affinity propagation clustering does not require the number of clusters to be predetermined (Frey & Dueck, 2007). Another proposed benefit of affinity propagation clustering is that it identifies exemplar observations for each cluster type (Brusco et al., 2019); however, other algorithms such as k ‐medoids also do this (Scrucca et al., 2016; Macqueen, 1967; Madhulatha, 2011). Affinity propagation clustering relies on what is known as a message‐passing algorithm that takes similarities between data points as the input (Brusco et al., 2019).…”

“…Another proposed benefit of affinity propagation clustering is that it identifies exemplar observations for each cluster type (Brusco et al., 2019); however, other algorithms such as k ‐medoids also do this (Scrucca et al., 2016; Macqueen, 1967; Madhulatha, 2011). Affinity propagation clustering relies on what is known as a message‐passing algorithm that takes similarities between data points as the input (Brusco et al., 2019). The data points can be considered as occurring in a network, and because of the structure of the network, affinity propagation clustering considers all data points simultaneously, which means that unlike other algorithms, the results will not be influenced by choosing the initial set of points (Dueck, 2009).…”

A Novel Approach for Cognitive Clustering of Parkinsonisms through Affinity Propagation

“…Cluster analysis is an unsupervised learning technique introduced by Tryon [1], which has an aim to partition objects into homogenous groups for detecting the natural structure as well as the underlying patterns of a dataset according to a measure of similarity, for example geometrical distance. Unlike supervised learning [2,3], clustering is totally data-driven and clustering methods such as traditional hierarchical [4,5] and non-hierarchical [6,7] represent a valuable support to the neuropsychologists [8][9][10] for clustering patients and discovering different patterns of cognitive impairment or multiple cognitive profiles within diagnostic groups [11][12][13][14][15]. Moreover, cluster analysis resulted to be a powerful tool for finding the intra-and inter-diagnostic heterogeneity of psychological and cognitive performance of participants with bipolar disorder or depression syndrome [11][12][13][14] or of patients with Mild Cognitive Impairment [15].…”

“…However, the vast majority of clustering approaches for understanding the natural structure of cognitive performance in the neurodegenerative diseases [17,18] are traditional methods such as K-means [25], K-median and Ward's [5], which are less effective, or even inapplicable, on certain kind of cognitive data [8]. Moreover, a crucial issue of these clustering algorithms is the initial choice of the optimum number of clusters for the specific dataset [26,27], which choice, in most applications, could not be based on a priori knowledge [8].…”

“…More in detail, the AP algorithm is based on passing messages between data points, where the value of the messages measures the current eligibility of a candidate point to serve as exemplar for another data point [28]. The AP algorithm showed efficiency in handling high dimensional problems [8,[28][29][30] and it was applied widely in the physical sciences, especially for the clustering of gene expression.…”