ZHANG Yijin, MA Rui, LIN Zongbing
High-dimensional data often contains a large number of redundant features,which significantly impact the efficiency of data mining and the generalization performance of machine learning algorithms.Dimensionality reduction is regarded as a crucial preprocessing step,which can enhance generalization performance and reduce computational costs.Feature selection is one of the predominant techniques for data dimensionality reduction.In this paper,we propose an unsupervised feature selection method based on hesitant fuzzy regularization and autoencoders.Firstly,two different projection methods are used to calculate the hesitant fuzzy correlation coefficient between features.Additionally,the absolute value of the Pearson correlation coefficient and the cosine correlation coefficient between features are computed.The correlation coefficient matrix of the features is then used to construct the regularization term.Secondly,the regularization term based on the correlation matrix is incorporated into the process of autoencoder feature selection to enhance the identification of redundant features.Finally,two different feature contribution measures are utilized to rank the features and select a subset of features.The objective function is optimized using back-propagation algorithm and proximal gradient descent method,and clustering and classification experiments are conducted over six representative datasets.The results of these experiments demonstrate the superiority of the proposed method.We also performs sensitivity analysis,stability analysis,and convergence analysis with the proposed method,showing its effectiveness.
