DocumentCode :
2181439
Title :
Model Selection in Support Vector Machines Using Self-Adaptive Genetic Algorithm
Author :
Zhang, Ying ; Li, Lijie
Author_Institution :
Ningbo Coll. of Health Sci., Ningbo, China
Volume :
1
fYear :
2010
fDate :
29-31 Oct. 2010
Firstpage :
114
Lastpage :
118
Abstract :
The support vector machine is a powerful supervised learning algorithm that has been successfully applied to a plenty of fields including text and image recognition, medical diagnosis and so on. The kernel and its parameters optimization, formally known as model selection, is a crucial factor which influences a good tradeoff between bias and variance. To automate model selection of support vector machine, this paper presents a strategy utilizing self-adaptive genetic algorithm and data distribution to determine the kernel function and all the free kernel parameters. The model selection criterion using a novel fitness function with hyperplane radius and a off-spring individual selection method during the process of constructing the model. The experiments on well known benchmark data sets are carried out to validate the effectiveness of proposed strategy. The experimental results show that using genetic algorithm to tune the model is a promising way to enhance the performance of support vector machine.
Keywords :
genetic algorithms; learning (artificial intelligence); self-adjusting systems; support vector machines; benchmark data sets; data distribution; free kernel parameters; hyperplane radius; image recognition; kernel function; medical diagnosis; model selection criterion; novel fitness function; off-spring individual selection method; parameters optimization; self-adaptive genetic algorithm; supervised learning algorithm; support vector machines; text recognition; Computational modeling; Data models; Equations; Genetics; Kernel; Mathematical model; Support vector machines; Genetic Algorithm; Model Selection; Support Vector Machine;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Computational Intelligence and Design (ISCID), 2010 International Symposium on
Conference_Location :
Hangzhou
Print_ISBN :
978-1-4244-8094-4
Type :
conf
DOI :
10.1109/ISCID.2010.45
Filename :
5692677
Link To Document :
بازگشت