Feature selection method based on Laplacian operator

Abstract

The invention discloses a feature selection method based on a Laplacian operator. The feature selection method not only takes the correlation between a sample and a class tag into account, but also keeps the interdependence relation among samples. Specifically, a proposed Lap-Lasso method comprises two regularization items, and the first regularization item is a sparse regularization item which ensures that only a small quantity of features can be selected. In addition, a new regularization item based on Laplacian is introduced and is used for keeping local adjacent structure information among samples of the same types. Furthermore, an APG namely an Accelerated Proximal Gradient algorithm is adopted to optimize a proposed model. An experimental result in a UCI data set verifies the validity of the Lap-lasso method.

Description

technical field [0001] The invention discloses a feature selection method based on a Laplacian operator, and relates to the technical field of machine learning algorithms. Background technique [0002] Traditional algorithms in machine learning often suffer from the well-known curse of dimensionality. In this case, reducing the dimensionality of data is beneficial to improve the efficiency and accuracy of data analysis. Feature selection is the process of selecting a subset of the most relevant features from a set of features to reduce the dimensionality of the feature space, so as to achieve the goal of improving the performance of the learning model. [0003] Researchers have proposed various feature selection methods. These methods broadly fall into two categories: (1) feature ranking methods; (2) feature subset search methods. The feature ranking method usually considers the importance of each feature separately and ranks it, so as to select a set of the most importan...

AI Technical Summary

Problems solved by technology

[0012] In these Lasso-based methods, one of the main shortcomings is that only the dependencies between samples and predicted values ​​(ie labels) are considered, while the interdependence between samples and samples is ignored, such as similar samples. The local adjacent structure, the loss of these information may affect the discriminability of the selected features, thus affecting the final classification performance of the classifier

Images