基于改进烟花算法的SVM特征选择和参数优化

Improved Fireworks Algorithm for Support Vector Machine Feature Selection and Parameters Optimization

本文提出了一种改进的基于烟花算法的SVM特征选择和参数优化算法.该算法针对特征选择问题的0-1特性,使用二进制编码的烟花算法,采用基于RBF核函数的SVM,在选取尽可能少的特征数目的同时提高了分类准确率.通过UCI数据仿真,对比结果表明:该方法避免了过早成熟而陷入局部最优的问题,可有效地找出合适的特征子集及SVM参数,并取得较好的分类效果.

In this paper, we propose a 1,1 ,-- selection and parameters optimization in training SVM. For the 0--1 characteristic of feature selection, the binary coding Fireworks Algorittgn and RBF kernel function based SVM are used to improve the accuracy of classification with less features. Compared to previous works, the proposed method can avoid being mature and falling into a local value, and it can effectively find the appropriate feature subset and parameters to get better performance of classification in UCI dataset.