Year: 2007
Type: book-chapter
Abstract: From the results obtained, we can conclude that although NDA obtains good results with non-linear projections in all data sets, similar results can be obtained using GLP in most of them. This indicates to us that, in practice, linear projections can obtain results just as good as non-linear projections in most cases. Even though PCA employs linear projections as well, it performs worse in some data sets probably because it is an unsupervised method. Classical linear, supervised feature extraction methods like LDA have important limitations: first, the number of new features is limited by the number of classes; and second, numerical problems arise when working with high dimensional or small data sets, restricting its use.
DOI: 10.5772/4972
Authors: Alberto J. Perez Jimenez, Juan Carlos Perez Cortes
Institution: Universitat Politècnica de València
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4211020535