Logo image
Data Visualization by PCA, LDA and ICA
Thesis

Data Visualization by PCA, LDA and ICA

Yang, Tsun Yu
Masters, 國立清華大學, 資訊系統與應用研究所
2014

Abstract

視覺化 主成份分析 線性識別分析 獨立成份分析 Data Visualization Principal Component Analysis Linear Discriminant Analysis Independent Component Analysis
While the internet applications are widely used, the amount of data information has a rapid growth. Big data analysis has become a popular study issue nowadays. Data visualization could help us deal with these big data in an intuitive way. We apply linear dimensionality reduction methods to project the observation data into lower-dimensional subspace. The method could preserve most of the data characteristic with omitting an acceptable little information. Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) are two common linear mapping methods which are proved to have an effective classification result. Independent Component Analysis (ICA) is originally proposed to solve the blind source separation problem. Same as a linear mapping method, ICA also computes a mapping matrix for data projection. But unlike PCA and LDA, ICA assumes non-Gaussian distributions of data could separate the original sources from a mixture. In the implementation of data visualization, we reduce the dimensionality to 2 in order to present the projected data on Euclidean geometry. The experiment result shows that LDA has a better data classification performance than PCA. An ICA algorithm has random factors in itself which may lead to sundry results. By means of low dimensional data visualization, one can imagine or reveal the structure of high dimensional data, for example, the characteristic of clustering.

Metrics

1 Record Views

Details

Logo image