2014 | OriginalPaper | Buchkapitel
Dimensionally Reduction: An Experimental Study
verfasst von : Xin Song, Zhihua Cai, Wei Xu
Erschienen in: Advanced Data Mining and Applications
Verlag: Springer International Publishing
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
Dimensionality reduction as the available method to overcome the “curses of dimensionality” has attracted wide attention. However, the pervious studies treat the visualization and the subsequent classification performance separately. In this case, we do not know whether there is a underlying relationship (
i.e.,
direct proportion) between visualization and the followed classification performance.
In this paper we compare several dimensionality reduction techniques on three different types of data sets: 1) Benchmark, 2) Image, and 3) Text data. Specifically, to intuitively evaluate the quality of the dimension reduced data, the visualization analysis is carried out, in which we use a covariance matrix related criteria to quantify the information content of the data. Moreover, we also consider the classification accuracy in different latent spaces (
i.e.,
dimensions) as another performance criterion to further analysis the relationship between the information content and the subsequent classification task. The experimental results show that there is no direct proportion relationship between the information content and the further classification performance.