A fuzzy twin support vector machines based on feature-weighted hybrid affiliations
The fuzzy twin support vector machine(FTSVM)ignores the differences between different features,resulting in the fact that the calculation of kernel function or distance cannot accurately reflect the similarity between the samples,which makes FTSVM fail to achieve good classification results when dealing with high-dimensional data classification problems containing a large number of irrelevant or weakly correlated features;and the design of the degree of affiliation does not ef-fectively distinguish the outliers or noise.To address the above problems,this paper proposes a fuzzy twin support vector ma-chine(FM-FTSVM)based on feature-weighted hybrid affiliation.Firstly,the information gain of each feature is calculated,and the weights are assigned to the features based on the magnitude of the information gain value,which reduces the role of irrelevant or weakly relevant features and enables them to be better applied to the classification problem of high-dimensional data;then,a minimum enclosing sphere is constructed for each class of samples to calculate the feature-weighted affiliation based on the closeness,and combined with the feature-weighted affiliation based on the distance to obtain the feature-weigh-ted hybrid affiliation,which is a combination of the feature-weighted affiliation based on the distance from sample points to the center of the class.The combined consideration of the feature-weighted Euclidean distance from the sample point to the class center and the degree of closeness between samples can better identify outliers or noisy data.Finally,the fusion fea-ture-weighted kernel function reduces the impact of irrelevant features on the calculation of the kernel function or distance.Compared with the algorithm on artificial dataset,high dimensional dataset and UCI dataset,the proposed algorithm in this paper has advantages in distinguishing outliers,noise and valid samples,and it can also achieve better classification effectin high dimensional data.
fuzzy twin support vector machinefeature weightinginformation gainaffinitymembershiphigh-dimensional data
国家科技期刊平台
NSTL
万方数据
