Data characteristic selection method with structure maintenance characteristic

Abstract

The invention discloses a data characteristic selection method with a structure maintenance characteristic. A more effective monitoring-free characteristic selection algorithm can be obtained, according to the algorithm, a self-expression model is used to model a characteristic selection problem,. noise problem caused by learning pseudo label data can be avoided, the robustness of the algorithm isimproved by adding the structure maintenance characteristic, and a clustering result of higher precision is obtained. The method is realized by the steps that (1) an original data set X is determined, and the self-expression model of the original data set X is constructed; (2) a local manifold structure maintenance constraint is added to the self-expression model; (3) a reconstruction coefficientmatrix W added with the local manifold structure maintenance constraint is constrained to obtain a target function expression; (4) the target function expression is solved in an optimized way; and (5) characteristic selection is carried out on a characteristic selection matrix obtained by solution.

Description

technical field [0001] The invention belongs to the technical field of information processing, and in particular relates to a data feature selection method with structure-preserving characteristics. Background technique [0002] Feature selection is a very effective data analysis technique. It has attracted the attention and research of many scholars and has achieved good results in many practical tasks. It has been applied to image processing and computer vision, such as face clustering. , handwritten character recognition and object classification. From the perspective of whether to use labeled data, feature selection can be divided into three categories: supervised feature selection, semi-supervised feature selection, and unsupervised feature selection. [0003] Supervised feature selection is trained on labeled data, and then experimentally verified on test samples. Because the supervised feature selection method uses a large amount of prior information, it can obtain h...

AI Technical Summary

Problems solved by technology

[0009] Although filtering-based methods are relatively easy to implement, such methods evaluate features separately and often ignore global information, so they cannot achieve better feature selection results on some tasks.

Encapsulation-based methods often have a large time complexity and are prone to overfitting, thereby reducing the result of feature selection

Embedding-based methods usually need to learn a class label matrix, but the real class label matrix is ​​composed of discrete values, which is difficult to solve. Therefore, current methods use continuous values ​​to approximate discrete values, which will introduce noise and get unstable Feature Selection Results

Images