Performance Evaluation of Kernels in Support
Vector Machine
Abstract -Recently, the Support Vector Machine (SVM)
algorithm becomes very common technique that developed for
pattern classification. This technique has been employed in many
fields such as bioinformatics and with different attributes of
datasetsfor instance numeric, nominal or mixed. One of the
significant issues that user faces when implementing the SVM is
choosing the appropriate kernel function with attributes of
datasetto be investigated. This paper studied the behavior of SVM
in regarding to the used attributes of dataset with different kernel
functions. It analyzed the influence of various datasets
descriptions on efficiency of (SVM)classification.SVM with these
kernels have been implemented in Matlab. The investigated kernel
functions are linear, polynomials, Sigmoid and Radial Based
Function (RBF) .
The evaluation process shows that the description of dataset
with the used kernel function affects the performance of SVM
classifier. Generally, SVM with linear and RBF achieved 100% in
classification process when Mushroom dataset is used, and 99%
when Sickle Cell Disease (SCD) is used.
Keywords: SVM classifier, Linear, Polynomials, Sigmoid,
Radial Based Function (RBF), Numeric, Nominal, Mixed.
I. INTRODUCTION
Support Vector Machine (SVM) is supervised
classification technique which is based on theory of statistical
learning developed by Vladimir N. [1]. SVM is the best
algorithm among algorithms that was developed for pattern
classification. Recently It has been adapted for other uses,
such as finding regression and distribution estimation. It used
in many fields such as bioinformatics, which is currently a
very active research area in many universities and research
institutes.
SVM plays significant role in classification process. The
process of analyzing dataset items according to their labels
named classification, where labels are the class information
of related data included in that datasets. Classification
technique is used in many research areas for instance
bioscience, economy and forecasting because it strengths the
decisions and summarize data which were collected
previously to be analyzed efficiently. Accordingly, using this
technique is not limited to computer scientist but also
economists, doctors [2][3][4].
The basic concept of SVM is based on binary classification as
it separates data points by straight line to classify the class
label. Whereas, in some datasets, it is not possible to use one
straight line to separate the data points. Kernel functions are
introduced to overcome the previous issue [5] [6].
This paper studies the performance of SVM method based on
its kernel functions by using various types of datasets. Four
different Kernel functions were investigated based on using
three datasets which having verity instances of classes. The
studied kernel functions are linear, polynomials, Sigmoid and
Radial Based Function. The main aim of this study is explain
the difference of SVM performance in term of used kernel
function and dataset. In addition to, compare the investigated
kernel function in term of: accuracy, recall and precision.
The reminder of this paper is structured as following: Section
2 presentes related work. Support Vector Machines (SVM) is
described in section 3. Dataset is described in section 4.
Section 5 illustrates the results. Conclusion is detailed in
section 6.
II. RELATED WORK
Few researches have been investigated the performance of
SVM in association with the known kernel functions. The
following describes some of these researches:
In [7] the authors reviewed the efficacy of support vector
machines regarding applied kernel functions in addition to
investigating the dimensionality approaches which are fixed
slope regressionand principal component analysis. The used
dataset was high-dimensional bio-imaging and the results
show that RBF is the best.
The Authors in [8] implemented SVM depending on different
four kernel-functions which are linear, polynomial, RBF and
sigmoid. The implementation were used the dataset of
spambase. The performance of SVM with all kernel functions
were evaluated using different parameters such as Support
Vector Classification SVC. The results show that the
performance of SVM was the best with linear kernel function
and SVC.
In [9] the authors investigated the performance of SVM when
using different kernel functions. This study has been
performed on Arabic alphabetic character images which
Intisar Shadeed Al-Mejibli
University of Information Technology and
Communications
Baghdad,Iraq
dr.intisar.almejibli@gmail.com
dr.intisar.almejibli@uoitc.edu.iq
Dhafar Hamed Abd
Al-Maaref University College
Al-Anbar, Iraq
Dhafar.dhafar@gmail.com
Jwan K.Alwan
University of Information
Technology and Communications
Baghdad, Iraq
jwanism@uoitc.edu.iq
Abubaker Jumaah Rabash
Department of Islamic Studies
Sunni Endowment Diwan
Al-Anbar, Iraq
Abj87r@gmail.com
96
2018 1st Annual International Conference on Information and Sciences (AICIS)
978-1-5386-9188-5/18/$31.00 ©2018 IEEE
DOI 10.1109/AiCIS.2018.00029