Particle swarm optimisation for integrating feature selection and feature construction on high dimensional data

Idheba Mohamad Ali O. Swesi (P72759)

Please use this identifier to cite or link to this item: https://ptsldigital.ukm.my/jspui/handle/123456789/513422

Full metadata record

DC Field	Value	Language
dc.contributor.advisor	Azuraliza Abu Bakar, Assoc. Prof. Dr.
dc.contributor.author	Idheba Mohamad Ali O. Swesi (P72759)
dc.date.accessioned	2023-10-16T04:36:30Z	-
dc.date.available	2023-10-16T04:36:30Z	-
dc.date.issued	2021-07-01
dc.identifier.other	ukmvital:130490
dc.identifier.uri	https://ptsldigital.ukm.my/jspui/handle/123456789/513422	-
dc.description	Classification on high-dimensional data with thousands of features is one of the most difficult challenges in machine learning tasks. The challenge is not only because of high dimensionality and small sample size, but also the presence of irrelevant and redundant features that may limit the performance of the classifier. Therefore, feature selection (FS) and feature construction (FC) are essential preprocessing techniques used to enhance the quality of the search space by reducing the dimensionality while improving the accuracy and efficiency. Although FS and FC have been studied by many researchers, applying them to high-dimensional data is still challenging due to the large search space. Evolutionary computation (EC) approaches have been proposed for feature selection and construction due to their global search ability. Among the many EC approaches, particle swarm optimisation (PSO) is a population-based method that continues to be of interest to researchers due to its lower computational cost and fast convergence rate. However, despite the achievement of PSO for FS and FC, researchers still need to overcome some weaknesses such as the stagnation in local optima. The PSO for feature construction also has a high computational cost, especially for highdimensional problems. In addition to these shortcomings, PSO is used for either feature selection or feature construction separately, whereas the combination of both tasks can achieve better performance than either of them. Therefore, this study intends to investigate new PSO approaches for both feature selection and feature construction on high dimensional data. First, the research aims to identify suitable parameter values of PSO. Secondly, it seeks to enhance the global best of PSO to find a better solution, and thus improve the search ability that can help the algorithm to avoid stagnation in local optima by incorporating a filter-based local search within the PSO. The goal has been achieved by proposing a new PSO for FS with filter-based local search method. Thirdly, this research further investigates the impact of applying clustering algorithms with PSO-based feature construction to narrow the search space, and to improve the classification performance of the constructed features. The aim has been achieved by proposing a cluster-based PSO for feature construction method. Finally, an integrated approach of feature selection and construction based PSO algorithm is investigated. The proposed algorithm is expected to improve classification performance and reduce the dimensionality. In this research, all the proposed methods are examined using 10 high dimensional datasets. Empirical experiments on these datasets showed that the improved PSO approaches outperforms the full features, as well as the original PSO and other well-known methods in terms of their classification performance and the number of features. The proposed approaches are designed to handle either feature selection, feature construction or a combination of both where the last combination achieved the best performance among them in most cases. This is because the selected features help to alleviate the overfitting problem that occurred in the constructed feature.,Ph.D
dc.language.iso	eng
dc.publisher	UKM, Bangi
dc.relation	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat
dc.rights	UKM
dc.subject	Universiti Kebangsaan Malaysia -- Dissertations
dc.subject	Dissertations, Academic -- Malaysia
dc.subject	High dimensional data
dc.subject	Particle swarm optimisation
dc.subject	Heuristic algorithms
dc.title	Particle swarm optimisation for integrating feature selection and feature construction on high dimensional data
dc.type	Theses
dc.format.pages	243
dc.identifier.barcode	005879(2021)(PL2)
Appears in Collections:	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Files in This Item:

File	Description	Size	Format
ukmvital_130490+Source01+Source010.PDF Restricted Access		2.63 MB	Adobe PDF	View/Open

Show simple item record Recommend this item