In principal component analysis (PCA), each pattern
is treated as a random vector of which each component is a
random variable with mean and variance
(37) |
(38) |
(39) |
Usually the joint probability density function of the random vector is unknown. In this case, the mean vector and covariance matrix of can be estimated by the method of maximum likelihood estimation (MLE) based on a set of observed data samples :
Note that the rank of the estimated covariance matrix is at most , due to the samples in the dataset , assumed to be are independent, and the additional constraint:
(41) |
The variance can be treated as the dynamic energy contained in , or the amount of information carried by , while the trace can be considered as the total amount of dynamic energy contained in . Also, the covariance can be considered as the mutual energy, a measure of the correlation between and . By normalizing the covariance , we get the correlation coefficient between and :
(42) |
Examples
Six normally distributed 2-D datasets are generated with zero mean and the following covariance matrices:
(43) |
These data points are plotted below, together with the correlation coefficient on top of each dataset.