Improved Search Strategies and Extensions to K-medoids-based Algorithms - Extended Report

In this paper a number of improvements are suggested that can be applied to most k-medoids-based algorithms. These can be divided into two categories conceptual / algorithmic improvements, and implementational improvements. These include the revisiting of the accepted cases for swap comparison and the application of partial distance searching and previous medoid indexing to clustering. We propose extensions to the problem of nearest neighbor search, by combining the previous medoid index with triangular inequality elimination and partial distance searching. An improved k-medoids algorithm using simulated annealing, CLASA, is also discussed, as is a novel mechanism for managing memory usage. Various hybrids of these search approaches are then applied to a number of k-medoids-based algorithms and we show that the method is generally applicable. For example, experimental results based on various datasets, including both artificial and real datasets, demonstrate that when applied to CLARANS the number of distance calculations can be reduced by up to 98% with similar average distance per object. Importantly, these search approaches can also be applied to nearest neighbor searching and other clustering algorithms.