基于Spark的肿瘤基因混合特征选择方法

Hybrid Feature Selection Method for Tumor Gene Based on Spark

为处理随微阵列技术发展而急剧增长的肿瘤基因数据,实现对肿瘤基因数据的特征选择,结合集成特征选择和混合特征选择,提出一种Spark分布式计算框架的混合特征选择方法。利用F-score特征选择方法去除无关特征,进行初步特征选择,结合F-score、多分类支持向量机递归消除法、基于随机森林的特征选择3种方法得到最优的特征子集,并采用支持向量机对特征子集进行分类预测。实验结果表明,该方法能通过选择较少的基因达到较高的分类准确率。

In order to deal with the tumor gene data which grows rapidly with the development of microarray technology,and achieve the feature selection of tumor gene data,combined with integrated feature selection and mixed feature selection,a hybrid feature selection method of Spark distributed computing framework is proposed.The F-score feature selection method is used to remove the extraneous features,and the preliminary feature selection is carried out.The optimal feature subsets are obtained by integrating F-score,multi-class support vector machine recursive elimination method and random forest based feature selection,and the feature subset is classified and predicted by support vector machine.Experimental results show that this method can select fewer genes to achieve higher classification accuracy.