Modeling Multivariate Functional Data using Generalized Canonical Correlation Analysis and Principal Component Analysis

Authors

  • Sohair Fahmy Higazi, Dina Hassan Abdel-Hady, Hani Ahmed Khedr

DOI:

https://doi.org/10.17762/msea.v71i4.1984

Abstract

The research aimed to study the methods of multivariate analysis Generalized Canonical Correlation Analysis (GCCA), and Principal Components Analysis (PCA) on the functional data, with the aim of finding the most suitable method in analyzing and modeling the functional data. This is done by conducting an applied study of actual data for measuring industry performance for a mobile phone industry company. Where the data included twenty-three variables, divided into six groups (latent variables); As well as the simulation study was applied on fifteen variables for a time series by controlling the form of the relationship between the variables that meets the requirements of the analysis, and these variables are included in three groups. In the actual study data, the search concluded that the GCCA was superior to the principal component method. Whereas the simulation study presented additional results indicating that it is not always possible to assert the superiority of GCCA over the PCA method. Where the simulation study indicated that the results depend on the nature of the correlation matrix of the relationship between the basic variables. If the relationship between the variables within each group is strong, and the interrelationship between the variables in the different groups is weak, it is preferable to perform modeling using the PCA method. The simulation study confirmed that, to model the data using GCCA, there must be activation variables between the groups. The experimental results showed that there are linked variables between the groups that activate the relationship between the latent variables. Whereas, if the relationships between all basic variables within and between groupsare similar, both methods give similar results.

Downloads

Published

2022-08-19

How to Cite

Sohair Fahmy Higazi, Dina Hassan Abdel-Hady, Hani Ahmed Khedr. (2022). Modeling Multivariate Functional Data using Generalized Canonical Correlation Analysis and Principal Component Analysis. Mathematical Statistician and Engineering Applications, 71(4), 10758–10783. https://doi.org/10.17762/msea.v71i4.1984

Issue

Section

Articles