Global Journal of Pure and Applied Mathematics.
ISSN 0973-1768 Volume 12, Number 6 (2016), pp. 5027-5038
© Research India Publications
http://www.ripublication.com/gjpam.htm
Variable Extractions using Principal Component
Analysis and Multiple Correspondence Analysis for
Large Number of Mixed Variables Classification
Problems
Hashibah Hamid
*1
, Nazrina Aziz
2
and Penny Ngu Ai Huong
3
1,2,3
School of Quantitative Sciences, UUM College of Arts and Sciences,
Universiti Utara Malaysia, 06010 UUM Sintok, Kedah Malaysia.
Abstract
Non-parametric smoothed location model is another powerful approach which
can be used to discriminate the objects that contain both continuous and binary
variables. However, the smoothed location model is infeasible in estimating
parameters when a large number of binary variables involved in the study. To
handle this issue, the combination of two variable extraction techniques
namely principal component analysis (PCA) and multiple correspondence
analysis (MCA) are carried out before the construction of the smoothed
location model. In fact, there are four types of MCA but only Indicator MCA
and joint correspondence analysis (JCA) will be discussed in this article. Thus,
the performance of the smoothed location model together with combination of
PCA and two types of MCA, i.e. Indicator MCA and JCA, will be compared
and evaluated. The overall results from simulation study show that the
smoothed location model performed better when the binary extraction is done
by JCA rather than the Indicator MCA in terms of misclassification rate and
computational efficiency.
Keywords: Smoothed location model, classification, principal component
analysis, Indicator MCA, JCA, variable extraction, large variables, mixed
variables.