1. UKM Learning and Research Repository
2. PHD Thesis / Tesis PHD
3. Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Title:	Hybrid harmony search algorithms for gene selection problems
Authors:	Salam Salameh Mahmoud Shreem (P54557)
Supervisor:	Salwani Abdullah, Assoc. Prof. Dr.
Keywords:	Universiti Kebangsaan Malaysia -- Dissertations
Issue Date:	16-Dec-2013
Description:	In microarray gene expression studies, selecting the smallest possible set of genes from microarray datasets for clinical diagnosis and prediction is one of the most difficult challenges in machine learning tasks. An efficient gene selection technique is needed to search for the most important subset of genes by eliminating spurious or non-predictive genes from the original dataset without sacrificing or decreasing the accuracy of the classification. Many researchers have attempted to address this problem by using either a filter or a wrapper approach. The filter approaches are computationally efficient but are independent of the induction algorithm. On the other hand, the wrapper approaches perform better than filter approaches but are computationally more expensive. Many meta-heuristic approaches have been proposed for gene selection problems due their efficiency in obtaining a better solutions in reasonable times. Among the many meta-heuristic approaches, the harmony search algorithm (HSA) is a recent population-based method that continues to be of interest to researchers because it is more flexible and has a well-balanced mechanism to improve both global and local exploration abilities. However, despite some progress in improving HSA approaches, researchers still need to overcome some shortcomings, such as slow convergence due to the HSA's fully random mechanism for generating the initial harmony memory. The HSA also has a large degree of diversification (higher exploration) but is weak in terms of its exploitation capabilities. Beside these drawbacks of the HSA, gene selection problems also present some challenges in terms of over-fitting. These issues have motivated the investigation of the HSA which, to date, has not been applied to gene selection problems. In order to tackle the abovementioned drawbacks, the research presented in this thesis provides hybrid filter and wrapper approaches for obtaining the most relevant subset of genes that leads to better classification accuracy, minimal selected number of genes, and computational time. The research firstly aims to investigate a new gene selection method that is based on the HSA and to identify suitable values for the parameters of the HSA. Secondly, it seeks to enhance the construction process of the initial harmony memory while satisfying the diversity of the solution by embedding a reference set within the HSA (coded as RSHSA). Thirdly, the research in this thesis further investigates the effect of integrating Symmetrical Uncertainty (SU) as a filter with the RSHSA as a wrapper (coded as SU-RSHSA) by employing their advantages in order to attain better classification accuracy. Finally, to speed up the convergence process of the SU-RSHSA, a Markov blanket local search is implemented to act as an intensification strategy (coded as SU-RSHSA-MB). The proposed approaches in this work are tested on 10 microarray datasets. The empirical study on these large-scale microarray gene expression datasets indicates that the HSA when enhanced with different modifications (i.e., RSHSA, SU-RSHSA, and SU-RSHSA-MB) outperforms the HSA and other best-known methods in the literature in terms of classification accuracy and minimal selected genes,Certification of Master's/Doctoral Thesis" is not available
Pages:	182
Call Number:	QH450.S554
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.