MATLAB Implementation of Affinity Propagation Clustering Algorithm with Practical Examples

Affinity Propagation (AP) clustering is a widely-used clustering algorithm that groups data points based on pairwise similarity measures without requiring pre-specification of cluster numbers. The algorithm operates by iteratively calculating two key matrices: the similarity matrix (which measures how well each point serves as an exemplar for others) and the responsibility/availability matrices (which determine cluster assignments). This implementation includes core MATLAB functions for handling the similarity computation, message passing between data points, and convergence checking. The code demonstrates how to initialize preference values, update responsibility and availability matrices through iterative message passing, and identify exemplars when the algorithm converges. Key features include automatic determination of cluster numbers based on data structure, handling of different similarity metrics, and visualization of clustering results. The provided examples show practical applications with real datasets, illustrating how to preprocess data, set appropriate damping factors to ensure convergence, and interpret clustering outcomes. The implementation also includes performance optimization techniques for handling large datasets efficiently through vectorized operations and matrix computations.