A Novel Parameter Optimization Algorithm Based on Immune Memory Clone Strategy

Introduction

The Support Vector Machine (SVM) is a new machine learning method that based on the Statistic Learning Theory (SLT) (Zhou & Yang, 2006; Bo, Yuchun, Yang-Qing, Chung-Dar, & Weber, 2005). The selection quality of SVM parameters and kernel functions has an effect on the learning and generation performance. In order to find the best parameters for SVM, many researchers have done a mass of study. The parameters in SVM are usually selected by man’s experience, such as n-folded cross-verification (Nello & John, 2006). Recently, there are some automatic parameter selection methods researched such as colony algorithm and genetic algorithm (Chunxiu, Huiren, & Chunxia, 2010; Xiangying, Huiyan, & Fengzhen, 2010; Ning, Zhigang, & Qi, 2009; Yuan & Guangchen, 2010). These methods are efficient and automatic for optimizing parameters in a certain degree. But they depend on optimization model construction, and convergence to local optimum sometimes. According to these problems, a parameters optimization method of SVM based on immune memory clone strategy (IMC) is proposed in this paper. The results of experiment show that the proposed method has more efficiency of optimization and higher accuracy rate of classification than other existent methods.

Parameters Optimization Algorithm of SVM Based on Immune Memory Clone Strategy

Support Vector Machine

SVM is based on the principle of structural risk minimization. The ideal of SVM is to search for an optimal hyperplane to separate the data with maximal margin. When the training set is nonlinear, the training vector x is mapped into a higher dimensional feature space by a nonlinear function,and in the feature space who’s dimension maybe infinite construct the optimal classification hyperplane and the classifier’s decision function. In order to construct the optimal hyperplane, the following optimization problem must be solved:

(1) Where * is inner product,are coefficient vector,,are slack variables and C is a penalty parameter to be chosen by user. Finally, the decision function as follows:

(2) Where are Support Vectors (SVs), Lagrange multipliers satisfy with , n is number of SVs,is bias value. Eq.(2) shows that kernel function and penalty parameter affect the performance of SVM (Hanbing & Yubo, 2011).