孪生支持向量机的特征选择研究

Research on feature selection of twin support vector machine

针对机器学习中数据分类的特征选择问题,提出了孪生支持向量机(Twin support vector machine,TWSVM)的另一种方法:LFTWSVM.首先求解TWSVM优化问题后将得到两个权重向量,先将这两个权重向量进行归一化处理,再把处理后的两个权重向量取绝对值相加,得到一个总权重向量,最后将总权重向量进行特征选择.通过实验,将得到的数据结果和TWSVM特征选择方法进行比较,LFTWSVM特征选择方法具有一定的优势.

Aiming at the feature selection problem of data classification in machine learning a new method of twin support vector machine（TWSVM）is proposed：LFTWSVM Firstly,two weight vectors can be gotten after the SVM optimization problem is solved.Then,these two weight vectors will be normalized,and be summed together with their absolute values.A total weight vector can be gotten and features will be selected from the total weight vector.The experiments show that the feature selection method in LFTWSVM has rather advantages compared with the TWSVM.