基于特征内相关和互信息的加权SVM算法

特征加权支持向量机没有考虑特征间的相关性,因此产生的冗余会形成干扰并对最后的分类结果产生负面影响。为解决这个问题,提出了一种基于特征内相关和互信息的特征加权算法,并将其应用于支持向量机。该算法引入了特征间相关系数作为衡量冗余度的一个指标,以此计算出惩罚因子,在特征加权向量机的基础上对权值进行处理,尽可能真实地体现出特征对分类的贡献度。经过多个数据集以及几种不同算法的实验比较,提出的新算法具有更好的鲁棒性和泛化能力。

Weighted Support Vector Machine Algorithm Based on Inner-correlations and Mutual Information of Features

The feature-weighted support vector machine (FWSVM) does not take into account the correlation among the features,thus the redundancy and the interference caused by it will have a negative impact on the final classification result.A feature weighting algorithm based on inner-feature correlation and mutual information was proposed and applied in support vector machines.The algorithm introduces the inter-feature coefficient as an index to measure the redundancy,and then calculates the penalty factor to deal with the weight on the basis of the feature weighting vector machine.Thus it realizes the contribution of the feature to the classification as much as possible.The comparison of experiments on multiple data sets with several different algorithms shows that the proposed new algorithm has better robustness and generalization ability.