Discriminant Canonical Correlation Analysis Of Time-Dependent Multivariate Data Structure

Abstract

Canonical Correlation Analysis (CCA), which is a widely used covariance analysis method is a technique that is not fundamentally designed for multivariate multiple time-dependent data (MMTDD) structure that could be suitably partitioned on two subsets of response and predictor variables. This means that for such data problems, the conventional CCA would not yield practical results. The literature also shows scanty work in this area. This study therefore designs and implements grouping scheme discriminant canonical correlation analysis (GSDCCA) for handling this problem so that the time effect is adequately captured in the computation of the correlation coefficient between the two sets of variables. It first identifies key matrices underlying the concert and presents the design in both theory and illustration. Using data on six weather conditions in Ghana spanning the period 2000 to 2021, the demonstrations show that correlation coefficient between heating and cooling sets of weather conditions varies at different time points, and that the overall correlations are quite higher than that obtained from data assumed to be time-independent. The procedure is therefore recommended as an innovative approach for handling MMTDD.