支持向量分类机的参数选择方法研究

Parameters Selection Method for Support Vector Classification

支持向量分类机(Support Vector Classification,SVC)的参数选择一直缺乏一种通用、完善的方法,很大程度上限制了它的应用。为解决SVC参数选择的难题,提出了一种基于启发式深度优先搜索(Heuristic Depth-first Search,HDFS)的SVC参数自动寻优方法。该方法将10-fold交叉验证的最大识别率作为目标,利用HDFS算法进行SVC参数寻优,减少了SVC的训练时间,提高了分类的精度,从而确保了SVC参数选择的准确性。将该算法用于3个基准数据集的仿真实验,结果表明该方法在保证分类精度前提下,大幅度缩短了训练建模时间,提高了运行效率,具有一定的推广意义。

There have been no perfect algorithms for the selection of the optimal parameters of support vector classification（SVC）,therefore,the applications of SVC are limite.In order to get optimal SVM parameter,a parameter selection method for SVC based on heuristic depth-first search（HDFS） is proposed in this paper.In this method,the ten-fold cross-validation recognition rate is used as the classification objection and HDFS is used for parameter selection,which can reduce the train time,improve the precision of SVC,and insure the accuracy of parameter selection.Results on 3 benchmark datasets show that the new method not only can assure the classification precision but also can reduce training time markedly.The new method has certain practical application significance.