Analysis of PCA/DCA and class-dependent DCA

Abstract

Dimension reduction is a powerful tool in handling large high dimension data – it allows for not only effective visualization but also efficient manipulation of the data. A popular tool, PCA (Principal Component Analysis) provides a way to reduce dimensions using solely the data points and not their labels. Using DCA (Discriminant Component Analysis), the data can be utilized to a fuller extent by incorporating class information into the dimension reduction process. Enhancing the use of class information, we can adjust DCA to use different dimension reduction schemes for different subsets of the data by class, creating a Class-dependent DCA system (CDCA). For this paper, we focus on using the three algorithms, PCA, DCA, and CDCA for visualization purposes in 2 or 3 dimensions.