Semi-supervised sparse least squares support vector machine based on Mahalanobis distance

To reflect the similarity of input samples and improve the sparsity of semi-supervised least squares support vector machine (SLSSVM), a novel semi-supervised sparse least squares support vector machine based on Mahalanobis distance (MS-SLSSVM) was proposed, which innovatively introduced the Mahalano...

Full description

Saved in:
Bibliographic Details
Published in:Applied intelligence (Dordrecht, Netherlands) Netherlands), 2022-09, Vol.52 (12), p.14294-14312
Main Authors: Cui, Li, Xia, Yingqing
Format: Article
Language:English
Subjects:
Algorithms
Artificial Intelligence
Clustering
Computer Science
Datasets
Kernels
Least squares
Linear transformations
Machine learning
Machines
Manufacturing
Mechanical Engineering
Optimization
Processes
Sparsity
Support vector machines
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:To reflect the similarity of input samples and improve the sparsity of semi-supervised least squares support vector machine (SLSSVM), a novel semi-supervised sparse least squares support vector machine based on Mahalanobis distance (MS-SLSSVM) was proposed, which innovatively introduced the Mahalanobis distance metric learning and the pruning method of geometric clustering into SLSSVM. MS-SLSSVM, which combines the advantages of metric learning and pruning, is a promising semi-supervised classification algorithm. For one thing, MS-SLSSVM effectively uses the Mahalanobis distance kernel to capture the internal mechanism of the two classes of input samples, in which Mahalanobis distance kernel solves the optimization problem with linear space to nonlinear space through simple linear transformation. For another, to solve the sparsity of SLSSVM, MS-SLSSVM can not only use geometric clustering pruning method to filter unlabeled samples in high-density regions, but also label high-confidence unlabeled samples of estimation error ranking. A series of experiments has been conducted on multiple datasets: artificial, UCI and realistic datasets. The experimental results prove the effectiveness of the proposed algorithm.
ISSN:0924-669X
1573-7497
DOI:10.1007/s10489-022-03166-3