基于Boruta-PSO-SVM的股票收益率研究

针对股票收益率的分类预测研究中支持向量机(SVM)存在的参数选择困难以及分类性能较差的问题,提出了一种基于特征选择(Boruta算法)和粒子群优化(PSO)算法SVM的新算法.通过Boruta算法对训练集进行特征选择,剔除无价值的特征以降低输入维度,同时引入PSO算法优化SVM核函数参数,从而提高SVM的分类性能.实验结果表明:相比决策树、神经网络及极限学习机算法,新算法取得了更高的分类精度,可以有效提高股票收益率的分类预测性能.

Research on stock yield based on Boruta-PSO-SVM

Research of stock yield has high reference value for stock investors.Aiming at the exiting parameters selection difficulties of support vector machines(SVM)and poor classification performance problems,an improved Boruta-particle swarm optimization(PSO)-SVM algorithm is put forward based on feature selection and PSO. Feature selection for training set to reduce input dimensions through Boruta algorithm.PSO is used to optimize the kernel function parameters of SVM,so as to improve the classification performance of SVM.Experimental results show that compare with decision tree,neural network,extreme learning machine algorithm,the new algorithm has higher classification precision,it can improve performance of classification prediction of stock yield.