=Paper=
{{Paper
|id=Vol-3331/paper01
|storemode=property
|title=Research on Classification Effect of PSO Optimized SVM
|pdfUrl=https://ceur-ws.org/Vol-3331/paper01.pdf
|volume=Vol-3331
|authors=Pengyu Zhou,Lei Liu,Yeguo Sun,Xing Zhao,Yihong Liu
|dblpUrl=https://dblp.org/rec/conf/ahpcai/ZhouLSZL22
}}
==Research on Classification Effect of PSO Optimized SVM==
https://ceur-ws.org/Vol-3331/paper01.pdf
Research on Classification Effect of PSO Optimized SVM 1
Pengyu Zhou1, Lei Liu1,*, Yeguo Sun2, Xing Zhao3, Yihong Liu1
1.
School of Computer Science, Huainan Normal University, Huainan, Anhui, 232038, China.
2.
School of Finance and Mathematics, Huainan Normal University, Huainan, Anhui, 232038, China.
3.
School of Chemistry and Materials Engineering, Huainan Normal, University, Huainan, Anhui, 232038,
China.
Abstract
Support vector machine (SVM) is a very good classification algorithm. When applying SVM,
it is necessary to select the corresponding kernel function and corresponding parameters.
Different choices will have a greater impact on the classification effect. In this paper, the
Gaussian kernel function is selected as an example, and the corresponding parameters are
penalty factor C and kernel function parameter g. The particle swarm optimization (PSO)
algorithm is used to find the optimal parameters C and g. Through simulation comparison, it
is found that the classification effect of the PSO optimized SVM algorithm is obviously
better than that of the original SVM algorithm itself. At the same time, this paper gives some
theoretical guidance for intelligent algorithm in optimizing SVM algorithm.
Keywords
SVM, machine learning, PSO
1. Introduction
Support vector machine (SVM) is a class of generalized linear classifiers that classify data in a
binary way according to supervised learning. Its decision boundary is the maximum margin
hyperplane for learning samples. SVM can also perform nonlinear classification through kernel
method, which is one of the common kernel learning methods. Because SVM classifier has the
advantages of good universality, high classification accuracy and fast classification speed, it is widely
used in agriculture, finance, medical industry and equipment fault diagnosis to classify and predict
data. Li Zhouzi et al. [1] used SVM model for face color recognition, Ma Lili et al. [2] used
PSO-SVM for the classification of optical fiber sensor network signals, Su Xiaojie et al. [3] studied
the SVM road extraction method based on SAR polarization characteristics, Pan Lipeng [4] used
SVM to classify English question bank topics, and Sun Peisheng et al. [5] used SOA-SVM model to
study photovoltaic array fault diagnosis, Kang Jiayu et al. [6] used ANOVA and BO-SVM to study
transformer fault diagnosis, Zuo Xueqian et al. [7] used PSO-LSSVM to predict the state of
centrifugal pump, Tang Yu et al. [8] used improved sparrow algorithm to optimize SVM to study
abnormal points, Fan Jie et al. [9] studied the prediction of natural gas hydrate formation conditions
based on PSO-SVM, and Wang Yushu et al. [10] studied the optimization of sandwich shaped charge
structure based on PSO and SVM. Relevant scholars have made good research results in different
areas of SVM research.
AHPCAI2022@2nd International Conference on Algorithms, High Performance Computing and Artificial Intelligence
EMAIL: *Corresponding author’s e-mail: 13605545615@163.com (Lei Liu)
© 2022 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
1
2. Correlation algorithm
2.1 Particle Swarm Optimization
Particle swarm optimization (PSO) is a common swarm intelligence algorithm, which is designed
to imitate the foraging behavior of birds. It starts to initialize a group of particles randomly, and then
finds the optimal solution through iterative optimization to achieve the ultimate goal. Particle swarm
optimization updates speed and location using the following formula:
vid (k + 1) = ω ∗ vid (k ) + c1r1 ( pid (k ) − xid (k )) + c2 r2 ( p gd (k ) − xid (k )) （1）
xid (k + 1) = xid (k ) + vid (k ) （2）
In the formula, ω is the inertia weight, c1 and c2 are learning factors, and the value range is [1, 2]
generally. r1 and r2 are random numbers, and the value range is [0, 1] generally. vid (k ) is the search
speed of the i particle in the d dimensions at time k , xid (k ) is the position of the i particle in
the d dimensions at time k , pid (k ) is the optimal position of the i particle in the d dimensions at
time k , and p gd (k ) is the optimal position of the i entire particle swarm in the d dimensions at time k .
In the process of studying particle swarm optimization, generally c1 and c2 , r1 and r2 , ω are the key
parameters to be optimized. The optimization performance of particle swarm optimization algorithm
often passes the typical single peak test functions: step and sphere functions; Classic multi peak
test functions: griewank and rastrigin functions are used for optimization performance test.
2.2 Support Vector Machine Algorithm
Support Vector Machine (SVM) is a binary classification model. SVM has been recognized by the
industry as the most successful and best performing algorithm in machine learning in the past decade.
It maps the feature vectors of instances to some points in space. The goal of SVM is to draw a line to
"best" distinguish the two types of points. The classification schematic diagram of SVM is shown in
Figure 1. The classification effect of SVM should achieve the minimum impact on the local
disturbance of samples and the strongest robustness of classification results.
Fig 1. Schematic diagram of SVM classification
When applying the SVM algorithm, the selection of kernel functions has a great impact on the
algorithm performance. Common kernel functions are:
2
（1）Linear kernel function
κ ( xi , x j ) = xiΤ x j
（2）Polynomial kernel function
κ ( xi , x j ) = ( xiΤ x j ) d
（3）Gaussian kernel function
|| xi − x j ||2
κ ( xi , x j ) = exp(− )
2δ 2
（4）Laplace kernel function
|| xi − x j ||
κ ( xi , x j ) = exp(− )
δ
（5） Sigmoid kernel function
κ ( xi , x j ) = tanh( β xiΤ x j + θ )
Among them, the Gaussian kernel function is the most commonly used kernel function. When the
Gaussian kernel function is selected, the setting of the penalty factor C ofSVM and the kernel function
parameter g has an important impact on the classification effect. Too large C will lead to over fitting,
too small C will lead to under fitting, too large g value will lead to too few support vectors, too small
g value will lead to more support vectors, and the number of support vectors will affect the speed of
training and prediction. The reasonable selection of C and g values can achieve better classification
efficiency and accuracy. The optimization function of particle swarm optimization algorithm can
realize the reasonable selection of parameters, so that SVM can achieve better classification effect.
3. Optimize the effect
3.1 SVM algorithm classification
The SVM algorithm is used to classify data sets. The kernel function of the SVM algorithm in this
paper is a Gaussian kernel function. The values of parameters C and g are 1 by default. The training
set contains 100 samples, and the test set contains 30 samples. The classification effect is shown in
Figure 2.
3.2 PSO optimized SVM algorithm classification
The PSO optimized SVM algorithm is used to classify the dataset. The selected kernel function is
also a Gaussian kernel function. The PSO algorithm is used to find the values of the parameters C and
g of the optimal SVM algorithm, and then the optimized SVM algorithm is used to classify the dataset.
The classification effect is shown in Figure 3. It can be seen from Figure 2 and Figure 3 that the
classification effect of the SVM algorithm optimized by the PSO algorithm is significantly improved
compared with the original SVM algorithm, which shows that the optimization effect of the PSO
algorithm can be well applied to the classification research of the SVM algorithm.
3
Fig 2. SVM classification Fig 3. PSO optimizes SVM classification effect
4. Conclusion
SVM is a very excellent classification algorithm, which has been widely used and improved by
many scholars and researchers can achieve good results. This paper combines the characteristics of
SVM algorithm classification, takes the selection of Gaussian kernel function as an example,
combines PSO algorithm to find the optimal parameters C and g of SVM algorithm, and then carries
out simulation classification on the dataset. The simulation results show that the classification effect
of the improved SVM algorithm using PSO is significantly higher than that of the original SVM
algorithm. Later, relevant research will continue to be carried out to optimize the SVM algorithm
through different intelligent algorithms to achieve better classification results.
5. Acknowledgments
This study received support from the Huainan Normal University Scientific Research Project
(Grant no. 2022XJZD030), and the Key Research and Development Plan Project Foundation of
Huainan (Grant no. 2021A2412), the University Natural Science Foundation of Anhui Province
(Grant no. KJ2021A0970).
6. References
[1] Li Z Z,Feng Y,Lin Z S et al.(2022)Face color recognition based on SVM model optimized by
improved PSO algorithm.Computer Simulation,4:241-247.
[2] Ma L L,Gao J,Shen Z J et al.(2022)Optical fiber sensor network data recognition system based on
PSO-SVM.LASER&INFRARED,5:734-739.
[3] Su X J,Liu X Q.(2022)Road extraction based on SAR polarization characteristics and
SVM.MICROELECTRONICS&COMPUTER,8:47-53.
[4] Pan L P.(2022)Research on Classification and Recommendation of English Item Bank Based on
SVM.Automation Technology and Application,8:140-142.
[5] SUN P S, CHEN T G, CHENG C et al.(2022)Research on Fault Diagnosis of Photovoltaic Array
Based on SOA-SVM Model.Journal of Power Supply,6:85-91.
[6] Kang J Y,Zhang S X,Zhang Q P et al.(2022)Fault Diagnosis Method of Transformer Based on
ANOVA and BO-SVM.High Voltage Engineering,9:1-16.
[7] Zuo X Q,Xiong Z,Nie L et al.(2022)Study of centrifugal pump fault prediction method based on
PSO optimization LS-SVM.Electronic Measurement Technology,3:51-56.
[8] Tang Y, Dai Q,Yang M Y et al.(2022)Improved Sparrow Algorithm to Optimize SVM Outlier
Detection.Computer Engineering & Science,15:98-107.
4
[9] Fan J, Xu X Y, Zhou S D et al.(2022)Prediction of natural gas hydrate formation conditions based
on PSO-SVM.Natural Gas Chemical Industry,47:1-6.
[10] Wang Y S，Guo R，Liu R Z et al.(2022)Optimum design of double layer shaped charge structure
based on PSO and SVM.Journal of Weapon Equipment Engineering,8:244-249.
5