Universiti Teknologi Malaysia Institutional Repository

Hybrid optimization for k-means clustering learning enhancement

Farhang, Yousef (2016) Hybrid optimization for k-means clustering learning enhancement. PhD thesis, Universiti Teknologi Malaysia, Faculty of Computing.

[img]
Preview
PDF
770kB

Official URL: http://dms.library.utm.my:8080/vital/access/manage...

Abstract

In recent years, combinational optimization issues are introduced as critical problems in clustering algorithms to partition data in a way that optimizes the performance of clustering. K-means algorithm is one of the famous and more popular clustering algorithms which can be simply implemented and it can easily solve the optimization issue with less extra information. But the problems associated with Kmeans algorithm are high error rate, high intra cluster distance and low accuracy. In this regard, researchers have worked to improve the problems computationally, creating efficient solutions that lead to better data analysis through the K-means clustering algorithm. The aim of this study is to improve the accuracy of the Kmeans algorithm using hybrid and meta-heuristic methods. To this end, a metaheuristic approach was proposed for the hybridization of K-means algorithm scheme. It obtained better results by developing a hybrid Genetic Algorithm-K-means (GAK- means) and a hybrid Partial Swarm Optimization-K-means (PSO-K-means) method. Finally, the meta-heuristic of Genetic Algorithm-Partial Swarm Optimization (GAPSO) and Partial Swarm Optimization-Genetic Algorithm (PSOGA) through the K-means algorithm were proposed. The study adopted a methodological approach to achieve the goal in three phases. First, it developed a hybrid GA-based K-means algorithm through a new crossover algorithm based on the range of attributes in order to decrease the number of errors and increase the accuracy rate. Then, a hybrid PSO-based K-means algorithm was mooted by a new calculation function based on the range of domain for decreasing intra-cluster distance and increasing the accuracy rate. Eventually, two meta-heuristic algorithms namely GAPSO-K-means and PSOGA-K-means algorithms were introduced by combining the proposed algorithms to increase the number of correct answers and improve the accuracy rate. The approach was evaluated using six integer standard data sets provided by the University of California Irvine (UCI). Findings confirmed that the hybrid optimization approach enhanced the performance of K-means clustering algorithm. Although both GA-K-means and PSO-K-means improved the result of K-means algorithm, GAPSO-K-means and PSOGA-K-means meta-heuristic algorithms outperformed the hybrid approaches. PSOGA-K-means resulted in 5%- 10% more accuracy for all data sets in comparison with other methods. The approach adopted in this study successfully increased the accuracy rate of the clustering analysis and decreased its error rate and intra-cluster distance.

Item Type:Thesis (PhD)
Additional Information:Thesis (Ph.D (Sains Komputer)) - Universiti Teknologi Malaysia, 2016; Supervisor : Prof. Dr. Siti Mariyam Shamsuddin
Uncontrolled Keywords:clustering algorithms, meta-heuristic methods
Subjects:Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Divisions:Computing
ID Code:78635
Deposited By: Widya Wahid
Deposited On:29 Aug 2018 07:53
Last Modified:29 Aug 2018 07:53

Repository Staff Only: item control page