摘要
SVM是一种基于核函数的机器学习算法,因为它具有良好的推广性和较好的性能,所以成为近些年来大家所关注的热点,但是该算法存在两个问题:一、如何提高SVM的计算精度;二、如何减少计算时间。本文提出一种使用免疫算子的SVM算法,该算法不但能够提高SVM的性能使其更加接近于实际问题,还能避免因问题太复杂使得结果不是最优解的情况。文中最后对样本进行了实验,结果说明了使用免疫算子的方法比经典方法在分类效果上有明显提高。
In recent years,the researches on SVM focus on two main areas. One is to improve the precision of the SVM algorithm,and another is to improve its speed. In this paper, a new method, which can appropriately tune multiple parameters in the kernel functions of SVM,is proposed. It cannot only improve the algorithm performance and make it approach to the real problem,but also avoid those methods available are too complex,the kernel must be differential and the result may be not optimal. Simulation results for data show the result based on this method is improved much more than normal algorithm's.
出处
《计算机科学》
CSCD
北大核心
2004年第2期109-110,119,共3页
Computer Science
基金
国家自然科学基金(60073053和60133010)