To verify the performance of the Fast PCA method in
terms of processing time and mean squared error we have
compared it with the performance of EVD based method
for PCA. For the processing time, the cputime2 is evaluated
while increasing the data dimensionality. Similarly, mean
squared error is evaluated while increasing the data dimensionality.
We have generated uniformly distributed random
vectors of dimensions starting from 100 and going up to
4000. The number of samples or feature vectors is fixed
and is 100. The dimension is reduced to h = 10 in all the
cases and the tolerance (Eq. (8)) is set to 0.01 in the verification
of the algorithm.