Reformulation of Direct Minimum/Maximum Autocorrelation Factors as a Generalised Eigenvalue Problem
Faculty of Computing, Health and Science
School of Engineering
Under the assumption of a two structure linear model of coregionalisation minimum/maximum autocorrelation factors (MAF) and direct minimum/maximum autocorrelation factors (DMAF) are linear transformation methods that decorrelate a set of attributes into uncorrelated factors for all lags. The usual procedure for determining the MAF/DMAF transformation matrix is via two successive principal component analyses. In this paper the MAF/DMAF transformation is reformulated and presented in a mathematically more tractable manner as the solution to the symmetric definite generalised eigenvalue problem. This elucidates the process underlying the transformation, that is, the simultaneous diagonalisation of two matrices by congruence.