基于Scikit Learn的SVM分类器算法优化

Improvement of SVM Classifier Algorithm Based on Scikit Learn

支持向量机(SVM)在高维度数据分类中表现出优异性能,可通过核函数对原始特征进行映射,解决原始空间线性不可分问题。但由于数据特征、维度不同,所以SVM在参数调整时,一般需要手动调整,效率较低且增加工作量。针对该问题,提出了一种基于Scikit Learn的SVM分类器参数调整优化方法。使用网格搜索对最优参数范围进行搜索,利用高斯径向基核函数进行参数调整,基于Python机器学习库Scikit Learn对不同参数、不同核函数的分类结果进行可视化观察,并在网格上显示其最优参数范围,寻找准确率高的参数分布。通过自动迭代的方式对参数进行更精确求解,设定相应值代入迭代计算。同时为防止陷入过拟合,设定最优参数邻域范围直接读取最优参数值。实验结果表明,所提出的方法可大量减少人工调参时间,且可以更精确地获得SVM的最优参数。

Support Vector Machine(SVM)classifier algorithm,which can map the original features by the kernel function to solve the linear inseparability problem in the original space,has excellent performance in high-dimensional data classification.However,due to the different data characteristics and dimensions,it is generally necessary to manually test the parameters of SVM with low efficiency and increased workload.To address this problem,this study proposes an optimization method for SVM classifier based on Scikit learning.The optimal parameter range is searched by grid search method,and the classification results of different parameters and different kernel functions are visualized in Scikit Learn of Python machine learning library.The optimal parameter range is displayed on the grid to find the parameter distribution with high accuracy.The parameters are solved more accurately by means of automatic iteration,and the corresponding values are set to be brought into the iteration calculation.Moreover,the neighborhood search of the optimal parameters is determined to choose the optimal parameters value directly,in order to avoid falling into over-fitting.Experimental results show that the time of manual parameter adjustment is reduced greatly and the accuracy of the optimal parameters obtained here is improved.