基于相似性特征聚类的加权无监督特征选择算法

Weighted unsupervised feature selection algorithm basedon similarity feature clustering

目前存在的无监督特征选择算法中往往会忽略特征与特征之间的关系,从而使得特征选择结果不理想。针对这个问题,提出了基于相似性特征聚类的加权无监督特征选择算法(Weighted unsupervised feature selection algorithm based on similarity feature clustering,WFSFC)。其基本思想是:给特征f的k个近邻赋以属性权重,从而定义了新的k近邻密度以及平均冗余度,然后根据新提出的的k近邻密度及平均冗余度选择聚类中心,将特征空间中剩余的特征按照类间距离最大,类内距离最小的原则分配到已经选好的聚类中心所在的类中,最后对所选特征进行k-means聚类,并对7个UCI数据集进行实验。实验结果表明,该算法的特征选择结果较优。

In the existing unsupervised feature selection algorithms,the relationship between features is often ignored,which makes the result of feature selection unsatisfactory.Aiming at this problem,this paper puts forward Weighted unsupervised feature selection algorithm based on similarity feature clustering(WFSFC).Its basic idea is:attribute weights were assigned to k neighbors of the characteric f,thus the new k neighbors'density and average redundancty were defined.Then the clustering center is selected according to the proposed k neighbors'density and average redundancy.In accordance with the principle of the largest distance between classes and the minimum distance within the class,the remaining features in the feature space were allocated to the class where the selected clustering center is.Finally,the selected features were clustered with k-means and seven UCI data sets were experienced.The results show that the algorithm has a better feature selection.