基于自适应遗传算法的SVM参数优化

A Parameters Optimization Method for an SVM Based on Adaptive Genetic Algorithm

摘要:

针对基于遗传算法支持向量机(SVM)训练时间较长以及分类精度较网格搜索法有所降低等问题,通过重新定义遗传算法参数的寻优范围,提出一种自适应遗传算法；该算法根据网格搜索法得到遗传算法参数的最佳寻优范围,然后遗传算法在这个范围内进行参数的精确寻优,最后得到分类的结果；这样不仅可以有效缩短训练时间,而且拥有更高的分类正确率；通过UCI中的10组经典数据集的实验结果可知,自适应遗传算法较之网格搜索法、 常规遗传算法、粒子群算法在训练时间上有较大的提升,并且拥有较高的分类准确率。

Abstract:

Aiming at that training needs a long time and classification accuracy worse than grid algorithm which reduced the benefits of support vector machine (SVM) based on genetic algorithm.This paper redefined the optimal range of the parameters of genetic algorithm,and proposed an adaptive genetic algorithm.According to the grid search method to get the best optimization genetic algorithm parameter range, and then the genetic algorithm in this range for accurate optimization of parameters, get the final classification results. It can not only shorten the time effectively, and it has higher classification accuracy.Experimental results on the UCI 10 classical dataset verify that adaptive genetic algorithm is better than the grid search method、traditional genetic algorithm、particle swarm optimization (PSO) algorithm in the aspect of training time,and has higher classification accuracy.