一种新的支持向量机核函数评估方法

摘要

To solve the problem of incomplete evaluation indicators and defective application back-ground of the traditional evaluation method for kernel of support vector machine (SVM),a new evaluation method for kernel of SVM is proposed.Firstly,the evaluation system based on similarity,stability,and complexity is established and the calculation criterion of each index is given.Then,the weight of each index is calculated through AHP and based on user requirements.At last,the weight matrices and score matrices are utilized to make the kernel comprehensive performance evaluation by comprehensive scores.The appli-cation results indicate that this method not only makes the kernel evaluation come true,but also provides a basis for the search and selection of the SVM.%针对传统支持向量机核函数评估中指标构建不完备、应用背景不强等问题,提出了一种新的支持向量机核函数评估方法.该方法首先构建了以相似性、稳定性、复杂性为评估指标的评估体系,给出了各指标的计算准则,再基于用户实际应用需求对各评估指标权重作层次分析,最后利用指标权重矩阵和指标评分矩阵以综合打分方式对核函数性能进行综合评估并排序.应用结果表明,该评估方法可行有效,不仅能够较好地实现对支持向量机核函数性能的分析和评估,也能为支持向量机的研究与选择提供依据.