Semi- supervised Affinity Propagation Clustering Algorithm Based on Mahalanobis Distance

Abstract

A semi-supervised affinity propagation clustering algorithm based on Mahalanobis distance （SAPBM） is proposed to try to solve some problems， including that the limitations of the distance measurement of the affinity propagation （AP） clustering algorithm and the low accuracy of the clustering algorithm for some complex data sets. Considering that the Mahalanobis distance is not affected by the sample dimension， in the similarity measurement of samples， the SAPBM algorithm replaces the Euclidean distance with the Mahalanobis distance， reducing the mutual interference between samples due to the influence of sample dimension； combining pairwise constraint information to improve the similarity between the data， so that the obtained similarity matrix can more accurately reflect the relationship between the data. Experiments are carried out on the UCI standard data set， and the experimental results show that the SAPBM algorithm has better clustering performance than the traditional AP clustering algorithm and the SAP clustering algorithm that only uses pairwise constraint information.

Tasks

Reproductions