基于核密度峰值聚类和细粒度噪声抑制的特征选择方法OA

In view of the problems that traditional feature selection methods are vulnerable to noise and the ob-tained feature space is prone to cause changes in data distribution,this study proposes a FNKC(Feature Selection based on Fine-Grained Noise Suppression and Kernel Density Peak Clustering).To overcome the influence of noise on feature selection,the possibility theorem of the PFCM(Possibilistic Fuzzy C-means Clustering Algorithm)is uti-lized in combination with the information particle criterion to extract data correlation,thereby improving the anti-noise performance.Introducing kernel density in high-dimensional space to measure clustering density can accurately cap-ture the spatial structure within clusters,thereby better reflecting the distribution of data.Finally,the superiority and effectiveness of the proposed algorithm are verified by comparing multiple advanced feature selection methods on six public high-dimensional datasets.