Improved Clustering and Soft Computing Algorithms

Author: Shu-Chuan Chu

Chu, Shu-Chuan, 2004 Improved Clustering and Soft Computing Algorithms, Flinders University, School of Informatics and Engineering

This electronic version is made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. This thesis may incorporate third party material which has been used by the author pursuant to Fair Dealing exceptions. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact copyright@flinders.edu.au with the details.

Abstract

Clustering algorithms have been widely applied and different approaches have been developed for different domains of application. This thesis investigates efficient and effective clustering and soft computing algorithms. In the investigation of k-medoids algorithms, several improved algorithms are proposed, such as the Clustering Large Applications Based on Simulated Annealing (CLASA) algorithm, the Multi-Centroids with Multi-Runs Sampling Scheme (MCMRS) and Incremental Multi-Centroid, Multi-Run Sampling Scheme (IMCMRS) algorithms. The Partial Distance Search (PDS), Triangular Inequality Elimination (TIE) and Previous Medoid Index are also presented to improve the clustering speed of k-medoids based algorithms. In addition, a new memory utilization scheme is derived and applied to efficient k-medoids algorithms. In the investigation of centroid-based clustering algorithms, the tabu search with simulated annealing algorithm is proposed and applied to codebook design for vector quantization. Genetic clustering is also presented for mean-residual vector quantization. Several theorems based on Hadamard Transform for nearest neighbour search are presented and applied to efficient cluster (codeword) search for vector quantization. A label bisecting clustering algorithm is proposed and applied to create a robust watermarking technique. Parallel particle swarm optimization based on three communication strategies are proposed to solve the problems in which the relationship between parameters are either independent, loosely correlated, strongly correlated or unknown. Seven communication strategies for Ant Colony Systems (ACS) are proposed to improve the ACS for the traveller salesman problem and the Constrained Ant Colony Optimization (CACO) based on the quadratic metric, sum of k nearest neighbour distance, constrained addition of pheromone and a shrinking range strategy is also proposed and demonstrated to be better than the Ant Colony Optimization with Different Favor (ACODF).

Keywords: Multi-Centroids with Multi-Runs Sampling Scheme (MCMRS) and Incremental Multi-Centroid,Multi-Run Sampling Scheme (IMCMRS),Clustering algorithms,k-medoids algorithsm,Clustering Large Applications Based on Simulated Annealing (CLASA) algorithm,the Multi-Centroids with Multi-Runs Sampling Scheme (MCMRS),Soft computing algorithms,Clustering algorithms,k-medoids algorithms
Subject: Statistical Science thesis, Computational Modelling thesis

Thesis type: Doctor of Philosophy
Completed: 2004
School: School of Computer Science, Engineering and Mathematics
Supervisor: John Roddick