基于蚁群算法的支持向量机参数选择方法研究

摘要

Support vector machine can solve the classification problem with small samples with nonlinear and high dimensions, which has strong practicability.However, the classification accuracy of SVM is significantly affected by its training parameter.At present, genetic algorithm and network optimization algorithm are common optimization algorithms for SVM.However, these methods are easy to relapse into local extremum, so that optimization result might be bad.In order to improve the performance of classifiers of SVM, this paper introduces ant colony algorithm to guide the selection of SVM model parameters.Thus ,the parameters selection problem of SVM can be considered as a compound optimization problem by setting the objective function.The ant colony algorithm was applied to search the value of optimal objective function with the fine performance of robustness and distributed computing.This method is compared with SVM model selection method based on GA method.The experiment result shows ant colony algorithm can get the optimization solution in shorter time and higher classification accuracy than GA.%研究支持向量参数选择优化问题,常用的支持向量机参数优化算法和遗传算法分别存在耗时长和易陷入局部最优值的缺陷,导致支持向量机的分类精度低.为了解决支持向量机参数优化问题,提出了基于蚁群算法的SVM分类器泛化方法.蚁群算法是一种优化搜索方法,具有较强的鲁棒性、优良的分布式计算机制,SVM参数的选取看作参数的组合优化,建立组合优化的目标函数,采用蚁群算法来搜索最优目标函数值.然后将方法与GA的SVM模型选择方法进行了比较.实验表明采用蚁群算法具有一定的优势,能在较短的时间内寻找到最优解,证明已改进的方法得到了最精确参数优化结果.