UUM Electronic Theses and Dissertation
UUM ETD | Universiti Utara Malaysian Electronic Theses and Dissertation
FAQs | Feedback | Search Tips | Sitemap

Hybrid ACO and SVM algorithm for pattern classification

Alwan, Hiba Basim (2013) Hybrid ACO and SVM algorithm for pattern classification. PhD. thesis, Universiti Utara Malaysia.

[thumbnail of s92846.pdf] Text
Restricted to Registered users only

Download (3MB)
[thumbnail of s92846_abstract.pdf]

Download (653kB) | Preview


Ant Colony Optimization (ACO) is a metaheuristic algorithm that can be used to
solve a variety of combinatorial optimization problems. A new direction for ACO is to optimize continuous and mixed (discrete and continuous) variables. Support Vector Machine (SVM) is a pattern classification approach originated from statistical approaches. However, SVM suffers two main problems which include feature subset selection and parameter tuning. Most approaches related to tuning SVM parameters discretize the continuous value of the parameters which will give a negative effect on the classification performance. This study presents four algorithms for tuning the
SVM parameters and selecting feature subset which improved SVM classification accuracy with smaller size of feature subset. This is achieved by performing the SVM parameters’ tuning and feature subset selection processes simultaneously. Hybridization algorithms between ACO and SVM techniques were proposed. The first two algorithms, ACOR-SVM and IACOR-SVM, tune the SVM parameters while
the second two algorithms, ACOMV-R-SVM and IACOMV-R-SVM, tune the SVM parameters and select the feature subset simultaneously. Ten benchmark datasets from University of California, Irvine, were used in the experiments to validate the performance of the proposed algorithms. Experimental results obtained from the proposed algorithms are better when compared with other approaches in terms of classification accuracy and size of the feature subset. The average classification
accuracies for the ACOR-SVM, IACOR-SVM, ACOMV-R and IACOMV-R algorithms are 94.73%, 95.86%, 97.37% and 98.1% respectively. The average size of feature subset is eight for the ACOR-SVM and IACOR-SVM algorithms and four for the ACOMV-R and IACOMV-R algorithms. This study contributes to a new direction for ACO that can deal with continuous and mixed-variable ACO.

Item Type: Thesis (PhD.)
Supervisor : Ku-Mahamud, Ku Ruhana
Item ID: 4419
Uncontrolled Keywords: Continuous ant colony optimization, Mixed-variable ant colony optimization, Support vector machine, Tuning SVM parameters, Feature subset selection.
Subjects: Q Science > QA Mathematics > QA71-90 Instruments and machines
Divisions: Awang Had Salleh Graduate School of Arts & Sciences
Date Deposited: 09 Mar 2015 12:15
Last Modified: 25 Apr 2016 01:17
Department: Awang Had Salleh Graduate School of Arts and Sciences
Name: Ku-Mahamud, Ku Ruhana
URI: https://etd.uum.edu.my/id/eprint/4419

Actions (login required)

View Item
View Item